{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:BSLIWLM7DJSEWAYMZSEVI6X7PQ","short_pith_number":"pith:BSLIWLM7","canonical_record":{"source":{"id":"2605.12574","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T12:04:11Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0223384ce36f0276d83e371bd0f4e226dbc3a2e6c4fac28fdb4a1988b1aa5060","abstract_canon_sha256":"f61dc55ee3d336d6a2ec2b18b44b5897eba8688396ef3e0ba958b603c8530d8e"},"schema_version":"1.0"},"canonical_sha256":"0c968b2d9f1a644b030ccc89547aff7c14a97d08c525e55511979d05fc29cc69","source":{"kind":"arxiv","id":"2605.12574","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12574","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12574v1","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12574","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"pith_short_12","alias_value":"BSLIWLM7DJSE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"BSLIWLM7DJSEWAYM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"BSLIWLM7","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:BSLIWLM7DJSEWAYMZSEVI6X7PQ","target":"record","payload":{"canonical_record":{"source":{"id":"2605.12574","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T12:04:11Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0223384ce36f0276d83e371bd0f4e226dbc3a2e6c4fac28fdb4a1988b1aa5060","abstract_canon_sha256":"f61dc55ee3d336d6a2ec2b18b44b5897eba8688396ef3e0ba958b603c8530d8e"},"schema_version":"1.0"},"canonical_sha256":"0c968b2d9f1a644b030ccc89547aff7c14a97d08c525e55511979d05fc29cc69","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:10:01.674086Z","signature_b64":"n9FwpPgpZ9zIjQYYTRLXRRGQpBZa9nJZqY1sNM+3jRZI3p2BT3Onr0RDU0A5eH4TxVu6dks4SUTvT8msxJ3sAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0c968b2d9f1a644b030ccc89547aff7c14a97d08c525e55511979d05fc29cc69","last_reissued_at":"2026-05-18T03:10:01.673329Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:10:01.673329Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.12574","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:10:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Kbzk5giR2nzgwKWN3CEXDzBXqPZxNBR5aABHXwB/C87xdO0wLBz8dPnrg3jBfuPgO5wMWiP2g0YRv5NGtKdADA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T17:24:49.546279Z"},"content_sha256":"a24fef28567e572bf549d71ce23396f67e065dea28c196f03cd57ded90d038cd","schema_version":"1.0","event_id":"sha256:a24fef28567e572bf549d71ce23396f67e065dea28c196f03cd57ded90d038cd"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:BSLIWLM7DJSEWAYMZSEVI6X7PQ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"DistractMIA: Black-Box Membership Inference on Vision-Language Models via Semantic Distraction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hongyi Tang, Yi Yang, Zhihao Zhu","submitted_at":"2026-05-12T12:04:11Z","abstract_excerpt":"Vision-language models (VLMs) are trained on large-scale image-text corpora that may contain private, copyrighted, or otherwise sensitive data, motivating membership inference as a tool for training-data auditing. This is especially challenging for deployed VLMs, where auditors typically observe only generated textual responses. Existing VLM membership inference attacks either rely on probability-level signals unavailable in such settings, or use mask-based semantic prediction tasks whose effectiveness depends on object-centric visual assumptions. To address these limitations, we propose Distr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments across multiple VLMs and benchmarks show that DistractMIA consistently outperforms both output-only and stronger-access baselines. Its performance on a medical benchmark further demonstrates applicability beyond object-centric natural images.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The core intuition that member samples remain more anchored to the original image semantics while non-member samples are more easily redirected toward the distractor; this assumption is invoked to motivate the design and is not independently verified in the abstract.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DistractMIA performs output-only black-box membership inference on vision-language models by inserting semantic distractors and measuring shifts in generated text responses.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ae43543e6332a0416279bbe950a9b1e05a11596d5933484a418a4b168bed0bec"},"source":{"id":"2605.12574","kind":"arxiv","version":1},"verdict":{"id":"8600fc85-7d2a-433e-b73d-c83fb597d044","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:39:46.499730Z","strongest_claim":"Experiments across multiple VLMs and benchmarks show that DistractMIA consistently outperforms both output-only and stronger-access baselines. Its performance on a medical benchmark further demonstrates applicability beyond object-centric natural images.","one_line_summary":"DistractMIA performs output-only black-box membership inference on vision-language models by inserting semantic distractors and measuring shifts in generated text responses.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The core intuition that member samples remain more anchored to the original image semantics while non-member samples are more easily redirected toward the distractor; this assumption is invoked to motivate the design and is not independently verified in the abstract.","pith_extraction_headline":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses."},"references":{"count":41,"sample":[{"doi":"","year":2022,"title":"Scaling up vision-language pre-training for image captioning","work_id":"de3b8f19-cf5a-44bb-9c54-35e4489c2ec1","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Prompt-rsvqa: Prompting visual context to a language model for remote sensing visual question answering","work_id":"67c883ec-30d4-4cb5-a36b-b479a4ead484","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Visionllm v2: An end-to-end generalist multimodal large language model for hundreds of vision-language tasks.Advances in Neural Information Processing Systems, 37:69925–69975, 2024","work_id":"05c179aa-d6d0-457a-99a7-1c987c8d462f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Rs5m and georsclip: A large- scale vision-language dataset and a large vision-language model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing, 62:1–23","work_id":"89b61771-b402-4044-9158-fc6ee3dff7c4","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"The Pile: An 800GB Dataset of Diverse Text for Language Modeling","work_id":"9b10667a-da61-4358-aceb-10578234d45d","ref_index":5,"cited_arxiv_id":"2101.00027","is_internal_anchor":true}],"resolved_work":41,"snapshot_sha256":"71c8bc82a4d23b7e5e7f7d500419421ff6ca5f029ca7a06c91b4253b00eb4bee","internal_anchors":6},"formal_canon":{"evidence_count":1,"snapshot_sha256":"5f377b3a542200061ff62c7be619683ab8659f03d3313376128352706356437b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"8600fc85-7d2a-433e-b73d-c83fb597d044"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:10:01Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"mEpLSs+3FXmBXROjZfI7+yyz8ZqSXlaWvot3pZLbjJIhk6BiICNBBy9tVkDd2xNymtMUQtQpFomiVWsj8BrrAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-31T17:24:49.547590Z"},"content_sha256":"94600fef6b0747237d144626227602136c0403cb68e02ef9bb5b0870bf34ae24","schema_version":"1.0","event_id":"sha256:94600fef6b0747237d144626227602136c0403cb68e02ef9bb5b0870bf34ae24"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/bundle.json","state_url":"https://pith.science/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-31T17:24:49Z","links":{"resolver":"https://pith.science/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ","bundle":"https://pith.science/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/bundle.json","state":"https://pith.science/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/BSLIWLM7DJSEWAYMZSEVI6X7PQ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:BSLIWLM7DJSEWAYMZSEVI6X7PQ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f61dc55ee3d336d6a2ec2b18b44b5897eba8688396ef3e0ba958b603c8530d8e","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T12:04:11Z","title_canon_sha256":"0223384ce36f0276d83e371bd0f4e226dbc3a2e6c4fac28fdb4a1988b1aa5060"},"schema_version":"1.0","source":{"id":"2605.12574","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.12574","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"arxiv_version","alias_value":"2605.12574v1","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.12574","created_at":"2026-05-18T03:10:01Z"},{"alias_kind":"pith_short_12","alias_value":"BSLIWLM7DJSE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"BSLIWLM7DJSEWAYM","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"BSLIWLM7","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:94600fef6b0747237d144626227602136c0403cb68e02ef9bb5b0870bf34ae24","target":"graph","created_at":"2026-05-18T03:10:01Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Experiments across multiple VLMs and benchmarks show that DistractMIA consistently outperforms both output-only and stronger-access baselines. Its performance on a medical benchmark further demonstrates applicability beyond object-centric natural images."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The core intuition that member samples remain more anchored to the original image semantics while non-member samples are more easily redirected toward the distractor; this assumption is invoked to motivate the design and is not independently verified in the abstract."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"DistractMIA performs output-only black-box membership inference on vision-language models by inserting semantic distractors and measuring shifts in generated text responses."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses."}],"snapshot_sha256":"ae43543e6332a0416279bbe950a9b1e05a11596d5933484a418a4b168bed0bec"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"5f377b3a542200061ff62c7be619683ab8659f03d3313376128352706356437b"},"paper":{"abstract_excerpt":"Vision-language models (VLMs) are trained on large-scale image-text corpora that may contain private, copyrighted, or otherwise sensitive data, motivating membership inference as a tool for training-data auditing. This is especially challenging for deployed VLMs, where auditors typically observe only generated textual responses. Existing VLM membership inference attacks either rely on probability-level signals unavailable in such settings, or use mask-based semantic prediction tasks whose effectiveness depends on object-centric visual assumptions. To address these limitations, we propose Distr","authors_text":"Hongyi Tang, Yi Yang, Zhihao Zhu","cross_cats":["cs.AI"],"headline":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T12:04:11Z","title":"DistractMIA: Black-Box Membership Inference on Vision-Language Models via Semantic Distraction"},"references":{"count":41,"internal_anchors":6,"resolved_work":41,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Scaling up vision-language pre-training for image captioning","work_id":"de3b8f19-cf5a-44bb-9c54-35e4489c2ec1","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Prompt-rsvqa: Prompting visual context to a language model for remote sensing visual question answering","work_id":"67c883ec-30d4-4cb5-a36b-b479a4ead484","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Visionllm v2: An end-to-end generalist multimodal large language model for hundreds of vision-language tasks.Advances in Neural Information Processing Systems, 37:69925–69975, 2024","work_id":"05c179aa-d6d0-457a-99a7-1c987c8d462f","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Rs5m and georsclip: A large- scale vision-language dataset and a large vision-language model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing, 62:1–23","work_id":"89b61771-b402-4044-9158-fc6ee3dff7c4","year":2024},{"cited_arxiv_id":"2101.00027","doi":"","is_internal_anchor":true,"ref_index":5,"title":"The Pile: An 800GB Dataset of Diverse Text for Language Modeling","work_id":"9b10667a-da61-4358-aceb-10578234d45d","year":2020}],"snapshot_sha256":"71c8bc82a4d23b7e5e7f7d500419421ff6ca5f029ca7a06c91b4253b00eb4bee"},"source":{"id":"2605.12574","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:39:46.499730Z","id":"8600fc85-7d2a-433e-b73d-c83fb597d044","model_set":{"reader":"grok-4.3"},"one_line_summary":"DistractMIA performs output-only black-box membership inference on vision-language models by inserting semantic distractors and measuring shifts in generated text responses.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"DistractMIA infers VLM training membership by adding a semantic distractor to an image and tracking shifts in generated text responses.","strongest_claim":"Experiments across multiple VLMs and benchmarks show that DistractMIA consistently outperforms both output-only and stronger-access baselines. Its performance on a medical benchmark further demonstrates applicability beyond object-centric natural images.","weakest_assumption":"The core intuition that member samples remain more anchored to the original image semantics while non-member samples are more easily redirected toward the distractor; this assumption is invoked to motivate the design and is not independently verified in the abstract."}},"verdict_id":"8600fc85-7d2a-433e-b73d-c83fb597d044"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a24fef28567e572bf549d71ce23396f67e065dea28c196f03cd57ded90d038cd","target":"record","created_at":"2026-05-18T03:10:01Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f61dc55ee3d336d6a2ec2b18b44b5897eba8688396ef3e0ba958b603c8530d8e","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-12T12:04:11Z","title_canon_sha256":"0223384ce36f0276d83e371bd0f4e226dbc3a2e6c4fac28fdb4a1988b1aa5060"},"schema_version":"1.0","source":{"id":"2605.12574","kind":"arxiv","version":1}},"canonical_sha256":"0c968b2d9f1a644b030ccc89547aff7c14a97d08c525e55511979d05fc29cc69","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0c968b2d9f1a644b030ccc89547aff7c14a97d08c525e55511979d05fc29cc69","first_computed_at":"2026-05-18T03:10:01.673329Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:10:01.673329Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"n9FwpPgpZ9zIjQYYTRLXRRGQpBZa9nJZqY1sNM+3jRZI3p2BT3Onr0RDU0A5eH4TxVu6dks4SUTvT8msxJ3sAg==","signature_status":"signed_v1","signed_at":"2026-05-18T03:10:01.674086Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.12574","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a24fef28567e572bf549d71ce23396f67e065dea28c196f03cd57ded90d038cd","sha256:94600fef6b0747237d144626227602136c0403cb68e02ef9bb5b0870bf34ae24"],"state_sha256":"b9463e960f6043a3f53fa2e4d07313b7713f71ff5e7c837117401be8b1cb14e3"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gakQSYT7L90RVcmODm8sJyaTL6ILa9JO93nQRs6x4VdDdtDHCi4VEOqifksr2XsA4OoGyApoYXvr+GVpJ6KRCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-31T17:24:49.553285Z","bundle_sha256":"f1a18bd53854c8fc300b5626db6140871d88ee5a278d0edc8616bdc219df6f14"}}