{"as_of":"2026-08-09T18:07:00Z","caps":{"database_statements":6,"inbound":100,"outbound":100},"context_digest":"sha256:bea140ecde691fb7a0ca48a6165a48bdcb75dac4e5ea1226bbbac65a25e9874d","coverage":[{"denominator":104,"lane":"reference_resolution","note":"Typed states for the displayed outbound observations.","records_observed":100,"source":"paper_references, paper_reference_links","source_observed_at":"2026-08-02T04:57:37.796578Z","state":"measured"},{"denominator":100,"lane":"standing_notices","note":"One-hop event checks from named stored sources.","records_observed":100,"source":"scholarly_work_events, retraction_status_cache","source_observed_at":"2026-08-09T06:31:02.800959+00:00","state":"measured"},{"denominator":0,"lane":"inbound_itemization","note":"Pith citing papers itemized under the disclosed page cap.","records_observed":0,"source":"paper_references, paper_reference_links","source_observed_at":null,"state":"measured"},{"denominator":1,"lane":"external_citation_measurements","note":"A source-named dated measurement, never combined with another source.","records_observed":0,"source":"cited_works","source_observed_at":null,"state":"measured"}],"external_citation_measurements":[],"inbound":[],"links":{"evidence":"/evidence","html":"/paper/2607.13541/citation-record","integrity":"/paper/2607.13541/integrity","json":"/paper/2607.13541/citation-record.json","paper":"/paper/2607.13541"},"outbound":[{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.010027Z","title":"Fake it till you make it: Learning transferable representa- tions from synthetic imagenet clones","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":1,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.010027Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c47ad3286b085bffa39654d64971d45cecd8514cb692e901cf722932d6a56638","observation_id":"99d373a0-027a-4394-ae6a-175c35fd8fe5","resolution":{"observed_at":"2026-08-02T04:57:30.010027Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.065780Z","title":"Deep residual learning for image recognition","venue":null,"work_id":null,"year":2016},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":2,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.065780Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:11e60fd725472ebf449b9c2930bc05f6fd0b6bf8bad3baf283dbffa2382542e7","observation_id":"3a93f91c-71da-4125-8c3d-22af5036b321","resolution":{"observed_at":"2026-08-02T04:57:30.065780Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.146147Z","title":"What chatgpt and generative ai mean for science.Nature, 614(7947):214–216, 2023","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":3,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.146147Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:82dc35e73654ef8581de59a42546bb98554167a82915b1c693a853ffbb992824","observation_id":"4e8adba3-d40f-4536-b1f4-68f50d2b41f2","resolution":{"observed_at":"2026-08-02T04:57:30.146147Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.228610Z","title":null,"venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":4,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.228610Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:27f3680ed047b7f560e99a2216d69d0df7cfd8a98c8ad6fb331c52122e679baf","observation_id":"9b6702e7-b65e-4d97-827a-7ebbce26062b","resolution":{"observed_at":"2026-08-02T04:57:30.228610Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.311903Z","title":"Real- fake: Effective training data synthesis through distribution matching","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":5,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.311903Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:dcf9acba4b70f6f9f4db00db55ea6da2fc1705d628561bcedc47f013b9a77260","observation_id":"11ec654e-58a8-4f1a-96b2-6c17c071f9bd","resolution":{"observed_at":"2026-08-02T04:57:30.311903Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.411638Z","title":"Is synthetic data all we need? benchmarking the robustness of models trained with synthetic images","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":6,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.411638Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:e90be1166c7f95fa7a2f3c32e3381273fc7911810ffd7693e68901f761c50240","observation_id":"262148d1-7a21-49cb-b637-3f38cbe693b5","resolution":{"observed_at":"2026-08-02T04:57:30.411638Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.478207Z","title":"Scaling laws of synthetic images for model training","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.478207Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:755e16491e00676e6849dece43495754ef942c2b1509e116a0530d2ed6ba88b4","observation_id":"220ebb28-21ff-4f4b-a27f-1f66560f5e94","resolution":{"observed_at":"2026-08-02T04:57:30.478207Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.530554Z","title":"High-resolution image synthesis with latent diffu- sion models","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":8,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.530554Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:de6c6ab6abc483de1015f7aaaf2ec37e1bd03c5c8294b6f7a71b75bd022696ca","observation_id":"bc782a61-4145-443d-9eeb-aa5dba19a8fe","resolution":{"observed_at":"2026-08-02T04:57:30.530554Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.586443Z","title":"Flux.1 kontext: Flow matching for in-context image generation and editing in latent space, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":9,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.586443Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:f7f9817d174d91adbdb89b538025f91afcd475321b39ec0853cea2ba13aafb6c","observation_id":"62da9937-8304-4edc-ae01-c1c68a9a23e5","resolution":{"observed_at":"2026-08-02T04:57:30.586443Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2507.06261","last_updated":"2025-12-19T14:25:46Z","snapshot_observed_at":"2026-07-06T02:11:23.670680Z","submitted_at":"2025-07-07T17:36:04Z","title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","version":6},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2507.06261","snapshot_observed_at":"2026-08-02T04:57:30.694774Z","title":"Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":10,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.694774Z"},"links":{"cited_paper":"/paper/2507.06261","citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:7432222cec8b4a53f9dee104b47cae62b5148ba204aab448481203d64252ae26","observation_id":"9cf28029-ba88-4672-a388-f9480596cb26","resolution":{"observed_at":"2026-08-02T04:57:30.694774Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.773566Z","title":null,"venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":11,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.773566Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:7a7346931b67e967784d83347c63f59fd10e47147cedd1cc15bd0c97a46d2248","observation_id":"8652b818-8eff-4210-bc1c-b0721af0a717","resolution":{"observed_at":"2026-08-02T04:57:30.773566Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:30.853983Z","title":"Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":12,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.853983Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:426635777cb703858c25cb3b20ab66daade731dca53350812c3b952587564c77","observation_id":"afc40952-4dfd-4f35-ae5e-1dcbe41a00f4","resolution":{"observed_at":"2026-08-02T04:57:30.853983Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2501.03575","last_updated":"2025-07-09T19:35:31Z","snapshot_observed_at":"2026-08-03T00:21:10.886100Z","submitted_at":"2025-01-07T06:55:50Z","title":"Cosmos World Foundation Model Platform for Physical AI","version":3},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2501.03575","snapshot_observed_at":"2026-08-02T04:57:30.959222Z","title":"Cosmos world foundation model platform for physical ai","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":13,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:30.959222Z"},"links":{"cited_paper":"/paper/2501.03575","citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:cca4fa21343f86aeb4a5917d13370b0ee2416398e3884f093222c3e193d8e186","observation_id":"bbb9bcc4-9b7b-4ea1-91b2-65c710a73a34","resolution":{"observed_at":"2026-08-02T04:57:30.959222Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.063341Z","title":"Does training with synthetic data truly protect privacy? InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":14,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.063341Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:a6aecce1bf68c3ec7432cc130537b4df7b19e1ff8cab07f06ce121faefa9281a","observation_id":"858d766d-dea2-4c80-a352-1953e1965e89","resolution":{"observed_at":"2026-08-02T04:57:31.063341Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.139090Z","title":"Membership inference attacks against machine learning models","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":15,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.139090Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:97834d8766ce14f8d09a5ca5d253f9a2aec1e40f9dd37fd0d607f16d3a2fc761","observation_id":"1c327ff9-c6e4-4461-8fc7-2c431062660c","resolution":{"observed_at":"2026-08-02T04:57:31.139090Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.239628Z","title":"Membership inference attacks from first principles","venue":null,"work_id":null,"year":1914},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":16,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.239628Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:a35c349b5a11d704fb85d293ceb04e211ca5ac183d452703608f289b6c69cf4f","observation_id":"166e684c-9581-4b88-9b72-dc31431c97e1","resolution":{"observed_at":"2026-08-02T04:57:31.239628Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.320852Z","title":"Comprehensive privacy analysis of deep learning: Passive and active white-box infer- ence attacks against centralized and federated learning","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":17,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.320852Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:b88a1ed53248f1286fd2240d2d5b97c574b0d042864fb2c37b5c24f73e50e4b0","observation_id":"0d220795-0f12-4d9c-800b-ebd650b7d297","resolution":{"observed_at":"2026-08-02T04:57:31.320852Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.401867Z","title":"Privacy risk in machine learning: Analyzing the connection to over- fitting","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":18,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.401867Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:129c8f81c33cfc2905787e937b0126981e837be93eb3c7b92d0c85dce38c9c0a","observation_id":"5e8a0dc7-25ad-4586-8d59-c278c92a07b2","resolution":{"observed_at":"2026-08-02T04:57:31.401867Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.481626Z","title":"Membership inference attacks as privacy tools: Reliability, disparity and ensemble","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":19,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.481626Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:23bb695c48ee22efa325b841febe03938adc270c6e449eb69590c4bbcc5127c1","observation_id":"f8926530-87ba-4962-9c42-024804a1c835","resolution":{"observed_at":"2026-08-02T04:57:31.481626Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.562695Z","title":"Gan-leaks: A taxonomy of membership inference attacks against generative models","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":20,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.562695Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:fb76eb240b683a18e71e32d8fd05e53eacb5a58fe3aab88b0f65c6d751f7ad10","observation_id":"5d87156c-1201-4f0e-92e9-445fd0237e5c","resolution":{"observed_at":"2026-08-02T04:57:31.562695Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.647702Z","title":"Enhanced label-only membership inference attacks with fewer queries","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":21,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.647702Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:da71df5931f0cb62ca6e023ea476e1adac48a64443f41ad2315395732e640d5e","observation_id":"610587c9-66f2-44ab-8ed2-6b52dd37e613","resolution":{"observed_at":"2026-08-02T04:57:31.647702Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.768887Z","title":"A method to facilitate member- ship inference attacks in deep learning models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":22,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.768887Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:16443d68a0a869a7852607f9ce074badec96334e1aa0cbad7f64ab97f4fe1f7d","observation_id":"f0839358-9972-40fc-98ce-154af4090878","resolution":{"observed_at":"2026-08-02T04:57:31.768887Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.774882Z","title":"Reconciling privacy and accuracy in ai for medical imaging","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":23,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.774882Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:0dfe9ae89dab89d52d0044a170666f7edfb286f0bb888b6fdba89d3d3c09b574","observation_id":"649b0bcd-543f-4ed1-93dc-e72723c7e447","resolution":{"observed_at":"2026-08-02T04:57:31.774882Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:31.909044Z","title":"Watermarking makes language models radioactive","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":24,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:31.909044Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:d7ce3db860e226c403e275a23aca07e3184867d71080099f46017877a1689ff2","observation_id":"77de4dff-afc3-4c11-8b93-ab67b959ad66","resolution":{"observed_at":"2026-08-02T04:57:31.909044Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.059247Z","title":"Any-resolution ai-generated image detection by spectral learning","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":25,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.059247Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:ecd6fb524bc31699325175f0588954a2359b0f19fd42898ca05c90bced018bbd","observation_id":"ff76adc2-1d75-47a0-94f8-50f994c5b034","resolution":{"observed_at":"2026-08-02T04:57:32.059247Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.189336Z","title":"Towards universal fake image detectors that generalize across generative models","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":26,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.189336Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c6ba477d5bf386d62a9d2ebb9a69c747c8c331e7205e0ff4a91768dfe232b2d1","observation_id":"54d721ca-2de7-4293-bfc7-45f426d05219","resolution":{"observed_at":"2026-08-02T04:57:32.189336Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.289001Z","title":"Membership inference attacks and de- fenses in neural network pruning","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":27,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.289001Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:b1db456733d57ed1933a2b6ca291de03b6befcefeccc650a14c396b25cbb2388","observation_id":"230631d5-c14e-456f-bb87-b68e37273d3b","resolution":{"observed_at":"2026-08-02T04:57:32.289001Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.407787Z","title":"Privacy risks of securing machine learning models against adversarial examples","venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":28,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.407787Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:7f06c70e2695be97b7c13e448659e8014a1ebbd947e4863d105fa3c9dd9e9c14","observation_id":"5dc662ac-3c89-4fd2-8f71-7d4fb6310091","resolution":{"observed_at":"2026-08-02T04:57:32.407787Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.525356Z","title":"Low-cost high-power membership inference attacks","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":29,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.525356Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c333e78aaaf8364819f11086df1c22bdd03d1e2d5c8c2c49da6910d106af4e42","observation_id":"d93252a0-f08e-4040-bb58-0ea447c33957","resolution":{"observed_at":"2026-08-02T04:57:32.525356Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.681563Z","title":"SDXL: Improving latent diffusion models for high-resolution image synthesis","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":30,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.681563Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:9610f20d0a859e4df1a28fa20fb889c9b2155040656937427bcbd1571374e40b","observation_id":"76840450-add2-4acd-9ca0-018c520b5cf7","resolution":{"observed_at":"2026-08-02T04:57:32.681563Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.843838Z","title":"Scalable diffusion models with transformers","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":31,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.843838Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c0b9e7561ee27e780561b13ec917ea0ebfb1e2517e91f0a60aa776ee9b669620","observation_id":"eb138729-f786-41d7-9fc4-44cceb58027c","resolution":{"observed_at":"2026-08-02T04:57:32.843838Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.950850Z","title":"SANA: Efficient high-resolution text-to-image synthesis with linear diffusion transformers","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":32,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.950850Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:392a5f704d9f92f628f9d22dfdf8b7317dede642df05ef6348b54757bb807d6d","observation_id":"9daf66e1-278c-49e1-beb1-3241e026d38b","resolution":{"observed_at":"2026-08-02T04:57:32.950850Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.953669Z","title":"LoRA: Low-rank adaptation of large language models","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":33,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.953669Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:735cc8feb2690958397422ad57fa981f246cbbde0f2ab0710a377d03a0668594","observation_id":"8af0fc2b-c3a5-47f8-8796-41953c19c7ad","resolution":{"observed_at":"2026-08-02T04:57:32.953669Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:32.959011Z","title":null,"venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":34,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:32.959011Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:af325726dcc34161ca1145b27718a8a2c93dfc619655c70490ee4bf39ea96b72","observation_id":"79938c94-2d4b-43cb-9ebd-102ea9c61730","resolution":{"observed_at":"2026-08-02T04:57:32.959011Z","resolver_source":null,"status":"parse_uncertain"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.020369Z","title":"Lens: Localization enhanced by nerf synthesis","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":35,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.020369Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:3c80cf46774bce88e098610f5dcc13cb447c6fabb2f93f04bd9dba23fc070e85","observation_id":"538c64ae-6e50-4f34-aae3-ee090169943f","resolution":{"observed_at":"2026-08-02T04:57:33.020369Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.102854Z","title":"Nerf-supervision: Learning dense object descriptors from neural radiance fields","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":36,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.102854Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:143458b0dce906d2ee3d73f74f9f4f8d3fb927773396995916b38ee708b2ba9a","observation_id":"dfb250b2-97da-4996-8d36-eba295542adf","resolution":{"observed_at":"2026-08-02T04:57:33.102854Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.174179Z","title":"Augmented reality meets computer vision: Efficient data generation for urban driving scenes.International Journal of Computer Vision, 126(9):961–972, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":37,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.174179Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:9c899e6ee42f1e7a2dafc42160d31e090a601ee93028967e4eed376dd93de80f","observation_id":"75244365-84fd-477e-93f3-64442891dd90","resolution":{"observed_at":"2026-08-02T04:57:33.174179Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.235807Z","title":"Learning deep object detectors from 3d models","venue":null,"work_id":null,"year":2015},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":38,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.235807Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:3569f71de846024c15ad8be9e6eaf0838e5eac81fa17a63149c886cab3d7830a","observation_id":"8d12a0cf-4383-4c28-a679-9b57f8f0f730","resolution":{"observed_at":"2026-08-02T04:57:33.235807Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"1811.10959","last_updated":"2020-02-24T23:25:50Z","snapshot_observed_at":"2026-07-06T07:17:20.813296Z","submitted_at":"2018-11-27T13:17:45Z","title":"Dataset Distillation","version":3},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"1811.10959","snapshot_observed_at":"2026-08-02T04:57:33.262320Z","title":"Dataset distillation.arXiv preprint arXiv:1811.10959, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":39,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.262320Z"},"links":{"cited_paper":"/paper/1811.10959","citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:28fba9ccc83a934238a2bacca98339e28babd68163526d26cd022e6a8b630a66","observation_id":"68a086b0-8104-4310-83f5-9a33a912fc6f","resolution":{"observed_at":"2026-08-02T04:57:33.262320Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.320424Z","title":"Dataset condensa- tion with gradient matching","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":40,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.320424Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:337e53f23095b4b8d517a9fae74b06e35f0958f6e68e8315c1fbcfd8411cc024","observation_id":"5243d175-9062-4143-b9c6-a2500afb055d","resolution":{"observed_at":"2026-08-02T04:57:33.320424Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.417231Z","title":"Dataset distillation by matching training trajectories","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":41,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.417231Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:eeb3e2f7cde4c1c1f9fff9fa1095401e7c10b12e6258e4b6342e9f464be1f97a","observation_id":"f05a058c-8d81-457c-9da3-ddd6fae43d9c","resolution":{"observed_at":"2026-08-02T04:57:33.417231Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.563912Z","title":"Privacy for free: How does dataset condensation help privacy? InInternational Conference on Machine Learning, pages 5378–5396","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":42,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.563912Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:1c4e77b7ed0e0b9b3ad2bf4026df2a0c0db6f64ccab50ccd5a4e60b3402990c3","observation_id":"2cc151a1-ab95-4f76-b567-81163437d0ae","resolution":{"observed_at":"2026-08-02T04:57:33.563912Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2209.14987","last_updated":"2022-09-29T17:50:23Z","snapshot_observed_at":"2026-08-09T10:27:38.564939Z","submitted_at":"2022-09-29T17:50:23Z","title":"No Free Lunch in \"Privacy for Free: How does Dataset Condensation Help Privacy\"","version":1},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2209.14987","snapshot_observed_at":"2026-08-02T04:57:33.724324Z","title":"privacy for free: How does dataset condensation help privacy","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":43,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.724324Z"},"links":{"cited_paper":"/paper/2209.14987","citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:4cadccfe8b69f3068b99225eaf101eaa89cbaf3bd85c3cdfbc8f9d1e7cd19b46","observation_id":"a04fc794-f577-4115-b29f-660610762687","resolution":{"observed_at":"2026-08-02T04:57:33.724324Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.758520Z","title":"Backdoor attacks against dataset distillation","venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":44,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.758520Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:5c17a981f52e4b69476e417c32146c2217f35ffcd9b63cce8670d34d4d857fa0","observation_id":"ce9eb20e-0375-41e7-9cdd-9faf0bee123f","resolution":{"observed_at":"2026-08-02T04:57:33.758520Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:33.833609Z","title":null,"venue":null,"work_id":null,"year":2023},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":45,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.833609Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:f3356245966c676916eb90dfcfa57830b593ef3193154ae66c3a240c38a06531","observation_id":"d67e36e2-6919-4370-8eab-56c9508af2f7","resolution":{"observed_at":"2026-08-02T04:57:33.833609Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2503.14734","last_updated":"2025-03-27T02:52:43Z","snapshot_observed_at":"2026-08-02T04:15:31.100670Z","submitted_at":"2025-03-18T21:06:21Z","title":"GR00T N1: An Open Foundation Model for Generalist Humanoid Robots","version":2},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2503.14734","snapshot_observed_at":"2026-08-02T04:57:33.913780Z","title":"Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":46,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:33.913780Z"},"links":{"cited_paper":"/paper/2503.14734","citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:7e749cd12747a31294735dc290a348b3ef2e19415ab4edae2df33a22dd498c89","observation_id":"21c26ff2-6014-46f5-aa8f-7a6bb073d409","resolution":{"observed_at":"2026-08-02T04:57:33.913780Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.020785Z","title":"Membership inference attacks and defenses in classification models","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":47,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.020785Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:42a109ff92a99c556a59ce9ba6af0fae10217e85245d8cd6bc243d821cb2fa50","observation_id":"1df9673d-890a-4928-9f9f-01637cd79a34","resolution":{"observed_at":"2026-08-02T04:57:34.020785Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.115568Z","title":"Rigging the foundation: Manipulating pre-training for advanced membership inference attacks","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":48,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.115568Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:ea4ddd743169f83bc81606ff91c5cb6550fda18ec5305c0f0c66618fa1de65c9","observation_id":"4c0e020f-c0cf-4e54-bb2e-ad9c21737690","resolution":{"observed_at":"2026-08-02T04:57:34.115568Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.158699Z","title":"Enhanced membership inference attacks against machine learning models","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":49,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.158699Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:8063e8a0e1b392c4db21b2826d6040b5e2162c92c33830c84b0ddaac042faf3d","observation_id":"0d585179-20af-4a41-a30b-d6470e95486b","resolution":{"observed_at":"2026-08-02T04:57:34.158699Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.206621Z","title":"Armanuzzaman, and Ziming Zhao","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":50,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.206621Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:d6d5858b59f0a3c29907f67dede6fd6b12a98b350c216e4ae80a79e78446affd","observation_id":"b511950a-b20e-4bee-97c2-8497c72e9708","resolution":{"observed_at":"2026-08-02T04:57:34.206621Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.250739Z","title":"Practical blind membership inference attack via differential comparisons","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":51,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.250739Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:3e8ba472f914d18a2c475c07a04847ce57ce56ae890b62fe1799b874447d8ffd","observation_id":"8a96f31b-19d4-4ea9-903a-c80d24f43e54","resolution":{"observed_at":"2026-08-02T04:57:34.250739Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.334354Z","title":"SOFT: selective data obfuscation for protecting LLM fine-tuning against membership inference attacks","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":52,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.334354Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:db142f968c3b2aeaf80e7dacf5be93f64f52ed3a66fe2289fda4c7b4910cfe5f","observation_id":"55577ac1-b42b-4e40-a844-7397408c1008","resolution":{"observed_at":"2026-08-02T04:57:34.334354Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.395565Z","title":"Querycheetah: Fast automated discovery of attribute inference attacks against query-based systems","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":53,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.395565Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:106f93db13b5b77547043eb10d4cbf494426184f3a1e781be622f33190a2d3f5","observation_id":"28a919c1-2c6c-4c03-bacc-cfd32820dca1","resolution":{"observed_at":"2026-08-02T04:57:34.395565Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.467672Z","title":"SLMIA-SR: speaker-level membership inference attacks against speaker recognition systems","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":54,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.467672Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:d4ec523836d5fa522e0dde2caf0ef67a31bdd81ebad648194b71fffb9f617d55","observation_id":"c0bd0adc-9240-4db8-b69b-cf1b9b996b59","resolution":{"observed_at":"2026-08-02T04:57:34.467672Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.535189Z","title":"Imitative membership inference attack","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":55,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.535189Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:f3ae15c955fe6d0ba56ca95892de2e9d5264a692c02d1e650e40c9b5ce80e26b","observation_id":"8aaf7bc9-a0a5-48c7-b050-646d4bf3af59","resolution":{"observed_at":"2026-08-02T04:57:34.535189Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.625312Z","title":"Cascading and proxy membership inference attacks","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":56,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.625312Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:d91dd2354aa05e38dca5aef8a439b6690140211514dd385ef73773f93b383ae1","observation_id":"e11fd00b-ec98-4680-a754-ad36142a21a9","resolution":{"observed_at":"2026-08-02T04:57:34.625312Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.710953Z","title":"Please tell me more: Privacy impact of explainability through the lens of membership inference attack","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":57,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.710953Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:d14cf5c11eb424e550d59078fe393037b90cde675261d6e7704d0bedd0d5ff4f","observation_id":"14e416e0-c0b1-48a9-a42c-5b7efe11b7ab","resolution":{"observed_at":"2026-08-02T04:57:34.710953Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.798852Z","title":"A unified membership inference method for visual self-supervised encoder via part-aware capability","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":58,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.798852Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:aa2166c621692ed452276e16af9621ed8599128da4dc0005fdaefe1247b570c2","observation_id":"f16d490c-1a16-49ac-a353-45c443c8a1b4","resolution":{"observed_at":"2026-08-02T04:57:34.798852Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.885064Z","title":"Encodermi: Membership inference against pre-trained encoders in con- trastive learning","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":59,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.885064Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c780e2ee0ceaa4bed1f4025a7b5dd90f74c15051c3998cb5c4c6e9bd703b168b","observation_id":"82b32d39-870b-462d-be19-5818973f4f64","resolution":{"observed_at":"2026-08-02T04:57:34.885064Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:34.973849Z","title":"When machine unlearning jeopardizes privacy","venue":null,"work_id":null,"year":2021},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":60,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:34.973849Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:835e2f768c5205dc8541d058abdddb9cbd2e4b75648885b0e1516b051e17af2b","observation_id":"d84813e8-534f-4c6f-b1fb-2073a422cc45","resolution":{"observed_at":"2026-08-02T04:57:34.973849Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.060536Z","title":"Compleak: Deep learning model compression exacerbates privacy leakage","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":61,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.060536Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:880b1505f70fcb06ebec89f2d86a2830520e9708e6a49b17ce07b7b04993099c","observation_id":"41fd2fc9-f98d-42bd-ad99-58e1ecb2ad3c","resolution":{"observed_at":"2026-08-02T04:57:35.060536Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.144991Z","title":"Riddle me this! stealthy membership inference for retrieval-augmented generation","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":62,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.144991Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:fc9b7d9f3843b85df00bde912972d79e1b1156186a7d453ab7271f387c5334d4","observation_id":"46c1e18b-d8d0-4fcd-85d3-a3caa1b238ba","resolution":{"observed_at":"2026-08-02T04:57:35.144991Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.222447Z","title":"InProceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security, CCS 2025, Taipei, Taiwan, October 13-17, 2025, pages 4184–4198","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":63,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.222447Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:02081c78a816b0ee25c8f1779e31e0fdd710615095a0e093fbb5a5c782a54299","observation_id":"2eace009-6602-4129-a067-8c94a5392d86","resolution":{"observed_at":"2026-08-02T04:57:35.222447Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.293049Z","title":"Membership inference attacks against vision-language models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":64,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.293049Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c2ca1e37fd37ba9f54bd83878d9297338cef03999f3f6881d5c451516028a3a5","observation_id":"10870001-1ebc-48ab-9602-e1fde4a940b7","resolution":{"observed_at":"2026-08-02T04:57:35.293049Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.377700Z","title":"Did the neurons read your book? document-level membership inference for large language models","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":65,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.377700Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:0ae09fc3bcdf536f5055f3c4af31e959f758d5248e7bdbc860f4616bb5486d2e","observation_id":"5bab84ce-04cf-4b5c-a690-c62b59c47b98","resolution":{"observed_at":"2026-08-02T04:57:35.377700Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.439176Z","title":"Membership inference attacks against in-context learning","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":66,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.439176Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:0e71cfa0f99fa9ded7e5970ecba21b22832bd5b4fabe31fac7f6875bb80fb5f6","observation_id":"a656117e-0f3e-4875-9229-c803cd043a1d","resolution":{"observed_at":"2026-08-02T04:57:35.439176Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.501004Z","title":"Towards label-only membership inference attack against pre-trained large language models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":67,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.501004Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:09ddcf731cd12b4d2d63b28ca01aef405075ff609b4067062a04e6a96c8bfadf","observation_id":"36ce452c-8f1d-4c82-a412-d2d32c139464","resolution":{"observed_at":"2026-08-02T04:57:35.501004Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.560562Z","title":"Membership inference attacks on tokenizers of large language models","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":68,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.560562Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:73580f61fbde7a11a836ff14136903ff2c277af3540b76927661e1867313331b","observation_id":"06f4d320-72f1-4bcf-b5b9-22d347ae70fa","resolution":{"observed_at":"2026-08-02T04:57:35.560562Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.595794Z","title":"Window-based membership inference attacks against fine-tuned large language models","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":69,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.595794Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:7a40ab142e562c2fa9b15c037caf51ee6df00680d1a5236d65997d41c73991be","observation_id":"7fa14882-134a-4f8f-a7ec-c0b3ac5361a8","resolution":{"observed_at":"2026-08-02T04:57:35.595794Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.660057Z","title":"Vidleaks: Membership inference attacks against text-to-video models","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":70,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.660057Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:ef936b4a0be6813a58d41ba71fe390905a818f4357b6ef8c1dd1425f8c9b7979","observation_id":"2befd7dc-9990-4da1-9b81-4deb60e52c12","resolution":{"observed_at":"2026-08-02T04:57:35.660057Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.729855Z","title":"Diffence: Fencing membership privacy with diffusion models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":71,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.729855Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:ec66b6f0494468dd7220cc79ccccc4965aec57f113a6983e124e341db7a94de9","observation_id":"c8010beb-2dbf-4773-bcf0-967ded80fddb","resolution":{"observed_at":"2026-08-02T04:57:35.729855Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.814269Z","title":"Black-box membership inference attacks against fine-tuned diffusion models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":72,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.814269Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:bb7d37754962654b4dd919f3c8292427749083b029b2b81088f632c59e29b0c4","observation_id":"a0575827-1f8f-40b2-937a-9d5d308a7813","resolution":{"observed_at":"2026-08-02T04:57:35.814269Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.897021Z","title":"Inference attacks against graph generative diffusion models","venue":null,"work_id":null,"year":2026},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":73,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.897021Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:bc1e17089922a95955db32f1d46e69b81ecf9e141395903da24ff22c8abb9c3a","observation_id":"9deac9f4-5442-4f94-93b1-e0f1ee488827","resolution":{"observed_at":"2026-08-02T04:57:35.897021Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:35.964714Z","title":"Visualizing data using t-sne.Journal of machine learning research, 9(11), 2008","venue":null,"work_id":null,"year":2008},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":74,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:35.964714Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:13236fc8ef050bf5fb56f83a43ff3401687d066518cc1bff98d6bb5818bef78b","observation_id":"cb58fb81-368e-4936-88c4-da41464fdac8","resolution":{"observed_at":"2026-08-02T04:57:35.964714Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.039818Z","title":"Does learning require memorization? a short tale about a long tail","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":75,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.039818Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:f066e72f1daf8d015ee36a673d4325d5610e14637da346a40b0be5c210e95499","observation_id":"c1ea287c-e76e-43ba-bdcb-8112890afc30","resolution":{"observed_at":"2026-08-02T04:57:36.039818Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.108192Z","title":"What neural networks memorize and why: Discovering the long tail via influence estimation.Advances in Neural Information Processing Systems, 33:2881–2891, 2020","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":76,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.108192Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:a2b1ecb5370fe1f91c6a3f7a84ba5e1f47bf53ebdb7d8bcc6c3b2a17ab401e52","observation_id":"12927411-a085-4ff5-a654-a76567d0ef36","resolution":{"observed_at":"2026-08-02T04:57:36.108192Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.198843Z","title":"A closer look at memorization in deep networks","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":77,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.198843Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:5904038f94002b45c19962654cc3ccd5bcd66da6af24363414a472b8901d52f0","observation_id":"e23e7b1e-9e4c-461c-bce8-6a28e99b8f80","resolution":{"observed_at":"2026-08-02T04:57:36.198843Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.295812Z","title":"Memorization through the lens of curvature of loss function around samples","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":78,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.295812Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:8412e96b644cd19a068ba02dc34925e06f942134e4b2765b03841c6cffde033c","observation_id":"48c1ea80-cece-4b92-ab18-e02a9cbac311","resolution":{"observed_at":"2026-08-02T04:57:36.295812Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.377594Z","title":"Seqmia: Sequential-metric based membership inference attack","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":79,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.377594Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:0b3be22cec494ee481efc1659526130a3981dac5defe377d18c80a669992ed18","observation_id":"9c03f369-1746-4900-8f1f-31d7f06da97c","resolution":{"observed_at":"2026-08-02T04:57:36.377594Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.489467Z","title":"Is difficulty calibration all we need? towards more practical membership inference attacks","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":80,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.489467Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:fa3392496e35ad8fd5443f182aa916e60c06c54cb5715076b6874fcc33c039d9","observation_id":"aaa4ba28-8389-4288-b8d0-7270367a33a6","resolution":{"observed_at":"2026-08-02T04:57:36.489467Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.587003Z","title":"Mem- bership inference attacks by exploiting loss trajectory","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":81,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.587003Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:56c76bd4b256fac43d4f2b10f8354a31adc61e5d68233c2d78ae7b75de9e0783","observation_id":"a959c678-ba50-4c0d-81ca-5cae3b5939a8","resolution":{"observed_at":"2026-08-02T04:57:36.587003Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.674933Z","title":"Watch out! simple horizontal class backdoor can trivially evade de- fense","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":82,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.674933Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:2a648de8369c19511dac913ee98de32afe173e651c49915e604a237826f13902","observation_id":"0bd5402d-7df3-4a36-8e81-de6c8a46d57b","resolution":{"observed_at":"2026-08-02T04:57:36.674933Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.743685Z","title":"Yes,{One- Bit-Flip}matters! universal{DNN}model inference depletion with runtime code fault injection","venue":null,"work_id":null,"year":2024},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":83,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.743685Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:f83c80a7e3774ce9dfadf917cb49da5a5969f843de641c1369acd61b3a5688f2","observation_id":"4f0faab7-52fd-4372-8d85-38dcf969fed2","resolution":{"observed_at":"2026-08-02T04:57:36.743685Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.799997Z","title":null,"venue":null,"work_id":null,"year":2019},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":85,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.799997Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:b311e58e938d092319a7728dc33a16f32575e791e5ffda18fed3c3b2bfa47367","observation_id":"cf12ab68-3923-439d-9cc1-6a197fb5a601","resolution":{"observed_at":"2026-08-02T04:57:36.799997Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.864047Z","title":"Patternnet: A benchmark dataset for performance evaluation of remote sensing image retrieval.ISPRS Journal of Photogrammetry and Remote Sensing, 145:197–209, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":86,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.864047Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:fe85f195bbfaa8668a107825c8e728ed3c1c6db5e867a55f4a609c4eb77d566c","observation_id":"df3ea377-234b-45ea-b9dc-d3cda6ad1ef1","resolution":{"observed_at":"2026-08-02T04:57:36.864047Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:36.932553Z","title":"Vggface2: A dataset for recognising faces across pose and age","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":87,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:36.932553Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:fb0e354dea6c1daa40fa10c932ea53dac54091a31d09744ba9203493cf6f98bd","observation_id":"ae89051e-486f-420a-a77a-ddb9a55e4596","resolution":{"observed_at":"2026-08-02T04:57:36.932553Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.001714Z","title":"Imagenet: A large-scale hierarchical image database","venue":null,"work_id":null,"year":2009},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":88,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.001714Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c3731622d3467b25f61a68602f388a174810b465056c894eac0fa12b5e63505f","observation_id":"902ef83b-204b-4ace-88dd-7ad318296d66","resolution":{"observed_at":"2026-08-02T04:57:37.001714Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.083926Z","title":"Contrastive multiview coding","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":89,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.083926Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:6ba765e85e0e4afa7a29ca154d4d85bee7e5538c72317f6ad8978799c1e37ae6","observation_id":"1f1e036b-fb4a-4bc4-bc99-ebf1e4d02370","resolution":{"observed_at":"2026-08-02T04:57:37.083926Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.140517Z","title":null,"venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":90,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.140517Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:725d68a742248cf61843b49e6fd38daece4097d48e178fe682b53aa6c21df00a","observation_id":"7c3ddd47-1bb4-4035-9dbd-453d63f8abc7","resolution":{"observed_at":"2026-08-02T04:57:37.140517Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.199852Z","title":"Black-box membership inference attacks against fine-tuned diffusion models","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":91,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.199852Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:11f373f749cfe2f454e398cad326571d0d2a1fd3b34c42e5e4c2c29596ef1688","observation_id":"a9e5edbc-e1fb-4020-8a58-dddeebb1dbb0","resolution":{"observed_at":"2026-08-02T04:57:37.199852Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.245673Z","title":"Towards reliable verification of unauthorized data usage in personalized text-to-image diffusion models","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":92,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.245673Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:bcee3bd3330ad6f0129f03162a905744f5ec857367417cba46998aab3bc61792","observation_id":"9cc54f15-b1bd-4477-b772-815f75e9849e","resolution":{"observed_at":"2026-08-02T04:57:37.245673Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.325687Z","title":"Pretender: Universal active defense against diffusion finetuning attacks","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":93,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.325687Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:a667e02d5ab32ddd887a64152ef4ab56896b319eaae08683887dd89da13aca94","observation_id":"1c32bc56-8f31-4a91-b10b-11e8cc3d98ae","resolution":{"observed_at":"2026-08-02T04:57:37.325687Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.381304Z","title":"Genomic privacy and limits of individual detection in a pool.Nature Genetics, 41(9):965–967, 2009","venue":null,"work_id":null,"year":2009},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":94,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.381304Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:b26b3e0aaa31d2ac72ad00584cb773050df28d4ad68cfacc9d3f1adc054853be","observation_id":"184ead63-cf96-43eb-ade7-6c32a2c9ae52","resolution":{"observed_at":"2026-08-02T04:57:37.381304Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.436071Z","title":"Prototypical networks for few-shot learning.Advances in Neural Information Processing Systems, 30, 2017","venue":null,"work_id":null,"year":2017},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":95,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.436071Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:68307f133f67db37b47b5ff9e7d352a87b076d70351b76b5eefad378d499dca5","observation_id":"90474e19-59dd-4dcd-b9da-f5cd8d4e4934","resolution":{"observed_at":"2026-08-02T04:57:37.436071Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.487084Z","title":"Prevalence of neural collapse during the terminal phase of deep learning training","venue":null,"work_id":null,"year":2020},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":96,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.487084Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:281adacadfbf1d5084faceed23b48b2ed820012a7b341a7cc8a8886f61faccf3","observation_id":"4a644696-7cae-40e6-b537-3be9e130e6b6","resolution":{"observed_at":"2026-08-02T04:57:37.487084Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.503675Z","title":"Neural collapse under mse loss: Proximity to and dynamics on the central path","venue":null,"work_id":null,"year":2022},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":97,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.503675Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:6faa98356d8c5fc9882aad75155e08e855f7940e32cc6ee8c4a0d473deac365c","observation_id":"93ba1e1f-583d-446a-9dfa-05cf5aa8dd12","resolution":{"observed_at":"2026-08-02T04:57:37.503675Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.525212Z","title":"Distance-based image classification: Generalizing to new classes at near-zero cost.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(11):2624–2637, 2013","venue":null,"work_id":null,"year":2013},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":98,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.525212Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:c9f58e717400fb3989df0fb19db05961e59788b2d6160bfb85c8ac003f825319","observation_id":"d57c3f81-597d-44d0-999d-b8c316f19d36","resolution":{"observed_at":"2026-08-02T04:57:37.525212Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.587542Z","title":"Cambridge University Press, 2018","venue":null,"work_id":null,"year":2018},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":99,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.587542Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:3145cb815f24339aa1201fd60e28e6c75b8227c4b50a1476f1befe3f0da5fd92","observation_id":"55faa7c3-c3f2-42e8-b831-4a451c1f1591","resolution":{"observed_at":"2026-08-02T04:57:37.587542Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.697398Z","title":"Calibrating noise to sensitivity in private data analysis","venue":null,"work_id":null,"year":2006},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":100,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.697398Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:12035c0ff229bc43edf089a985f75ff847e28990426bf40a92502d9a89ddc8d9","observation_id":"eeec5a0c-a397-4942-9545-ef4889d8f1d5","resolution":{"observed_at":"2026-08-02T04:57:37.697398Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":null,"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":null,"snapshot_observed_at":"2026-08-02T04:57:37.796578Z","title":"Deep learning with differential privacy","venue":null,"work_id":null,"year":2016},"citing_paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training","version":1},"reference_index":101,"source":"pdf_text","source_observed_at":"2026-08-02T04:57:37.796578Z"},"links":{"citing_paper":"/paper/2607.13541"},"observation_digest":"sha256:a89a6e19975c1fc8574430fa6339439d1feb274383948cf134088b7d61b09d96","observation_id":"fedb3970-4ec8-4ea4-85ce-058345607283","resolution":{"observed_at":"2026-08-02T04:57:37.796578Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}}],"paper":{"arxiv_id":"2607.13541","last_updated":"2026-07-15T07:44:14Z","latest_version":1,"primary_category":"cs.CR","snapshot_observed_at":"2026-08-09T10:28:17.006420Z","submitted_at":"2026-07-15T07:44:14Z","title":"When T2I Synthetic Data Backfires: Amplified Privacy Risks in Real-Synthetic Mix Training"},"reference_resolution":{"displayed":100,"state_counts":{"malformed_identifier":0,"metadata_mismatch":0,"parse_uncertain":1,"unresolved":99,"verified_exact":0,"verified_fuzzy":0},"total_outbound_references":104},"refusal":"A citation records a reference. It does not transfer a finding from one paper to another.","schema":"pith.paper-citation-record.v1","standing_sources":[{"observed_at":"2026-08-09T06:31:02.800959+00:00","source":"crossref"},{"observed_at":"2026-08-09T06:30:57.326959+00:00","source":"retraction_watch"}],"thesis":"As of 9 August 2026, this Paper Citation Record lists 100 of 104 outbound references and 0 inbound Pith citation observations for arXiv:2607.13541."}