{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:PMFQUSM6676QULD4UWSSX7G3VM","short_pith_number":"pith:PMFQUSM6","schema_version":"1.0","canonical_sha256":"7b0b0a499ef7fd0a2c7ca5a52bfcdbab015d86f2b945d91f3d1ad367643d8fe5","source":{"kind":"arxiv","id":"2307.15593","version":3},"attestation_state":"computed","paper":{"title":"Robust Distortion-free Watermarks for Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"John Thickstun, Percy Liang, Rohith Kuditipudi, Tatsunori Hashimoto","submitted_at":"2023-07-28T14:52:08Z","abstract_excerpt":"We propose a methodology for planting watermarks in text from an autoregressive language model that are robust to perturbations without changing the distribution over text up to a certain maximum generation budget. We generate watermarked text by mapping a sequence of random numbers -- which we compute using a randomized watermark key -- to a sample from the language model. To detect watermarked text, any party who knows the key can align the text to the random number sequence. We instantiate our watermark methodology with two sampling schemes: inverse transform sampling and exponential minimu"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2307.15593","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-07-28T14:52:08Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"5a15c4e56b7c848adf515399f2d2bb51c73086e06db5a818bb7a3da47f4adaab","abstract_canon_sha256":"fc8542de9e688e1fd80b14fe4612d397364072fcf6c1f51526dd9307418b2f42"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:01.718475Z","signature_b64":"jbVTv5ZipbfZ3Hb9YGcbluN+P9ojIMb7koAiCvtQr99rzSvyfT+83ItAhXCppkhuaRyW8YwfDw6TO9r5DQjuCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7b0b0a499ef7fd0a2c7ca5a52bfcdbab015d86f2b945d91f3d1ad367643d8fe5","last_reissued_at":"2026-07-05T08:28:01.718003Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:01.718003Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robust Distortion-free Watermarks for Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"John Thickstun, Percy Liang, Rohith Kuditipudi, Tatsunori Hashimoto","submitted_at":"2023-07-28T14:52:08Z","abstract_excerpt":"We propose a methodology for planting watermarks in text from an autoregressive language model that are robust to perturbations without changing the distribution over text up to a certain maximum generation budget. We generate watermarked text by mapping a sequence of random numbers -- which we compute using a randomized watermark key -- to a sample from the language model. To detect watermarked text, any party who knows the key can align the text to the random number sequence. We instantiate our watermark methodology with two sampling schemes: inverse transform sampling and exponential minimu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.15593","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2307.15593/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2307.15593","created_at":"2026-07-05T08:28:01.718058+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.15593v3","created_at":"2026-07-05T08:28:01.718058+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.15593","created_at":"2026-07-05T08:28:01.718058+00:00"},{"alias_kind":"pith_short_12","alias_value":"PMFQUSM6676Q","created_at":"2026-07-05T08:28:01.718058+00:00"},{"alias_kind":"pith_short_16","alias_value":"PMFQUSM6676QULD4","created_at":"2026-07-05T08:28:01.718058+00:00"},{"alias_kind":"pith_short_8","alias_value":"PMFQUSM6","created_at":"2026-07-05T08:28:01.718058+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":26,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.05694","citing_title":"Beyond Heuristic Tuning: Power-Calibrated LLM Watermarking","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2607.08400","citing_title":"TRACE: A Two-Channel Robust Attribution Watermark via Complementary Embeddings for LLM-Agent Trajectories","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2606.10099","citing_title":"Unsupervised Style Representation Learning for AI-Text Detection via Paraphrase Inversion","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00348","citing_title":"Block-wise Codeword Embedding for Reliable Multi-bit Text Watermarking","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25796","citing_title":"SAMark: A Self-Anchored Text Watermarking with Paragraph-Level Paraphrase Robustness","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18430","citing_title":"Signature filtering: a lightweight enhancement for statistical watermark detection in large language models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07235","citing_title":"ArcMark: Distortion-Free Multi-Byte LLM Watermark via Optimal Transport","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2404.02138","citing_title":"Topic-Based Watermarks for Large Language Models","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21002","citing_title":"Verifiable Provenance and Watermarking for Generative AI: An Evidentiary Framework for International Operational Law and Domestic Courts","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16336","citing_title":"Detecting Verbatim LLM Copy-Paste in Homework","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2303.11156","citing_title":"Can AI-Generated Text be Reliably Detected?","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2508.11548","citing_title":"Copyright Protection for Large Language Models: A Survey of Methods, Challenges, and Trends","ref_index":78,"is_internal_anchor":false},{"citing_arxiv_id":"2509.20924","citing_title":"RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18333","citing_title":"Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2601.02602","citing_title":"SWaRL: Safeguard Code Watermarking via Reinforcement Learning","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08826","citing_title":"Fundamental Trade-Offs in Multi-Bit Watermarking of Stochastic Processes","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08964","citing_title":"Trustworthy AI: Ensuring Reliability and Accountability from Models to Agents","ref_index":93,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05443","citing_title":"SLAM: Structural Linguistic Activation Marking for Language Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04305","citing_title":"SWAN: Semantic Watermarking with Abstract Meaning Representation","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00348","citing_title":"Block-wise Codeword Embedding for Reliable Multi-bit Text Watermarking","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12051","citing_title":"Can we Watermark Low-Entropy LLM Outputs?","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11546","citing_title":"RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08759","citing_title":"Optimal Multi-bit Generative Watermarking Schemes Under Worst-Case False-Alarm Constraints","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06662","citing_title":"Towards Robust Content Watermarking Against Removal and Forgery Attacks","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06987","citing_title":"Response Time Enhances Alignment with Heterogeneous Preferences","ref_index":102,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM","json":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM.json","graph_json":"https://pith.science/api/pith-number/PMFQUSM6676QULD4UWSSX7G3VM/graph.json","events_json":"https://pith.science/api/pith-number/PMFQUSM6676QULD4UWSSX7G3VM/events.json","paper":"https://pith.science/paper/PMFQUSM6"},"agent_actions":{"view_html":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM","download_json":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM.json","view_paper":"https://pith.science/paper/PMFQUSM6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.15593&json=true","fetch_graph":"https://pith.science/api/pith-number/PMFQUSM6676QULD4UWSSX7G3VM/graph.json","fetch_events":"https://pith.science/api/pith-number/PMFQUSM6676QULD4UWSSX7G3VM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM/action/storage_attestation","attest_author":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM/action/author_attestation","sign_citation":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM/action/citation_signature","submit_replication":"https://pith.science/pith/PMFQUSM6676QULD4UWSSX7G3VM/action/replication_record"}},"created_at":"2026-07-05T08:28:01.718058+00:00","updated_at":"2026-07-05T08:28:01.718058+00:00"}