{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:AIUESTTRHWC5CIIRP6V7Z5ADVR","short_pith_number":"pith:AIUESTTR","schema_version":"1.0","canonical_sha256":"0228494e713d85d121117fabfcf403ac69b6f93a7bcc405dbc8f28d75165e1b6","source":{"kind":"arxiv","id":"2306.04634","version":4},"attestation_state":"computed","paper":{"title":"On the Reliability of Watermarks for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Aniruddha Saha, John Kirchenbauer, Jonas Geiping, Kasun Fernando, Kezhi Kong, Khalid Saifullah, Manli Shu, Micah Goldblum, Tom Goldstein, Yuxin Wen","submitted_at":"2023-06-07T17:58:48Z","abstract_excerpt":"As LLMs become commonplace, machine-generated text has the potential to flood the internet with spam, social media bots, and valueless content. Watermarking is a simple and effective strategy for mitigating such harms by enabling the detection and documentation of LLM-generated text. Yet a crucial question remains: How reliable is watermarking in realistic settings in the wild? There, watermarked text may be modified to suit a user's needs, or entirely rewritten to avoid detection. We study the robustness of watermarked text after it is re-written by humans, paraphrased by a non-watermarked LL"},"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":"2306.04634","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-06-07T17:58:48Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"c3e0ae3eaec26001cd7fe9fcaef00fc3f51a787865e9b14d397107030c1b2e97","abstract_canon_sha256":"728427e7f0b132798b192d32e22797ecc6a96aa5a38cd524f5710732023033bf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:14:17.914627Z","signature_b64":"LsP7TbfEjNQY67XuSLhrlB0gqnAzXu7yvo+EFErybWantfRrJYunkrU+Iu7tRTr+Tc48gKPBczH8OsFPDc4gAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0228494e713d85d121117fabfcf403ac69b6f93a7bcc405dbc8f28d75165e1b6","last_reissued_at":"2026-07-05T08:14:17.914200Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:14:17.914200Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Reliability of Watermarks for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Aniruddha Saha, John Kirchenbauer, Jonas Geiping, Kasun Fernando, Kezhi Kong, Khalid Saifullah, Manli Shu, Micah Goldblum, Tom Goldstein, Yuxin Wen","submitted_at":"2023-06-07T17:58:48Z","abstract_excerpt":"As LLMs become commonplace, machine-generated text has the potential to flood the internet with spam, social media bots, and valueless content. Watermarking is a simple and effective strategy for mitigating such harms by enabling the detection and documentation of LLM-generated text. Yet a crucial question remains: How reliable is watermarking in realistic settings in the wild? There, watermarked text may be modified to suit a user's needs, or entirely rewritten to avoid detection. We study the robustness of watermarked text after it is re-written by humans, paraphrased by a non-watermarked LL"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.04634","kind":"arxiv","version":4},"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/2306.04634/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":"2306.04634","created_at":"2026-07-05T08:14:17.914247+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.04634v4","created_at":"2026-07-05T08:14:17.914247+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.04634","created_at":"2026-07-05T08:14:17.914247+00:00"},{"alias_kind":"pith_short_12","alias_value":"AIUESTTRHWC5","created_at":"2026-07-05T08:14:17.914247+00:00"},{"alias_kind":"pith_short_16","alias_value":"AIUESTTRHWC5CIIR","created_at":"2026-07-05T08:14:17.914247+00:00"},{"alias_kind":"pith_short_8","alias_value":"AIUESTTR","created_at":"2026-07-05T08:14:17.914247+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05694","citing_title":"Beyond Heuristic Tuning: Power-Calibrated LLM Watermarking","ref_index":53,"is_internal_anchor":true},{"citing_arxiv_id":"2606.31602","citing_title":"Robust Text Watermarking for Large Language Models via Dual Semantic Embeddings","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31602","citing_title":"Robust Text Watermarking for Large Language Models via Dual Semantic Embeddings","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25796","citing_title":"SAMark: A Self-Anchored Text Watermarking with Paragraph-Level Paraphrase Robustness","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30501","citing_title":"Linear Ensembles Wash Away Watermarks: On the Fragility of Distributional Perturbations in LLMs","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07235","citing_title":"ArcMark: Distortion-Free Multi-Byte LLM Watermark via Optimal Transport","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2402.06922","citing_title":"Whispers in the Machine: Confidentiality in Agentic Systems","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2507.07871","citing_title":"Mitigating Watermark Forgery in Generative Models via Randomized Key Selection","ref_index":22,"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":13,"is_internal_anchor":false},{"citing_arxiv_id":"2308.03825","citing_title":"\"Do Anything Now\": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16363","citing_title":"CSF: Black-box Fingerprinting via Compositional Semantics for Text-to-Image Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13095","citing_title":"Watermarking Should Be Treated as a Monitoring Primitive","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12529","citing_title":"BackFlush: Knowledge-Free Backdoor Detection and Elimination with Watermark Preservation in Large Language Models","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13095","citing_title":"Watermarking Should Be Treated as a Monitoring Primitive","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2305.19118","citing_title":"Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2309.00614","citing_title":"Baseline Defenses for Adversarial Attacks Against Aligned Language Models","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10611","citing_title":"Re-Triggering Safeguards within LLMs for Jailbreak Detection","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08964","citing_title":"Trustworthy AI: Ensuring Reliability and Accountability from Models to Agents","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04305","citing_title":"SWAN: Semantic Watermarking with Abstract Meaning Representation","ref_index":52,"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":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06662","citing_title":"Towards Robust Content Watermarking Against Removal and Forgery Attacks","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06865","citing_title":"Dataset Watermarking for Closed LLMs with Provable Detection","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR","json":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR.json","graph_json":"https://pith.science/api/pith-number/AIUESTTRHWC5CIIRP6V7Z5ADVR/graph.json","events_json":"https://pith.science/api/pith-number/AIUESTTRHWC5CIIRP6V7Z5ADVR/events.json","paper":"https://pith.science/paper/AIUESTTR"},"agent_actions":{"view_html":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR","download_json":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR.json","view_paper":"https://pith.science/paper/AIUESTTR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.04634&json=true","fetch_graph":"https://pith.science/api/pith-number/AIUESTTRHWC5CIIRP6V7Z5ADVR/graph.json","fetch_events":"https://pith.science/api/pith-number/AIUESTTRHWC5CIIRP6V7Z5ADVR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR/action/storage_attestation","attest_author":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR/action/author_attestation","sign_citation":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR/action/citation_signature","submit_replication":"https://pith.science/pith/AIUESTTRHWC5CIIRP6V7Z5ADVR/action/replication_record"}},"created_at":"2026-07-05T08:14:17.914247+00:00","updated_at":"2026-07-05T08:14:17.914247+00:00"}