{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:543QWPOG7D6S7A7GPDXTKK7BUV","short_pith_number":"pith:543QWPOG","schema_version":"1.0","canonical_sha256":"ef370b3dc6f8fd2f83e678ef352be1a56bff1f451bc52f702f9e0d6429714746","source":{"kind":"arxiv","id":"2502.18608","version":2},"attestation_state":"computed","paper":{"title":"Breaking Distortion-free Watermarks in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Dung Daniel T. Ngo, Guang Cheng, Hengzhi He, Manuela Veloso, Niccol\\`o Dalmasso, Saheed Obitayo, Shayleen Reynolds, Vamsi K. Potluru","submitted_at":"2025-02-25T19:52:55Z","abstract_excerpt":"In recent years, LLM watermarking has emerged as an attractive safeguard against AI-generated content, with promising applications in many real-world domains. However, there are growing concerns that the current LLM watermarking schemes are vulnerable to expert adversaries wishing to reverse-engineer the watermarking mechanisms. Prior work in breaking or stealing LLM watermarks mainly focuses on the distribution-modifying algorithm of Kirchenbauer et al. (2023), which perturbs the logit vector before sampling. In this work, we focus on reverse-engineering the other prominent LLM watermarking s"},"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":"2502.18608","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-02-25T19:52:55Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"92e4bbd504147b16abca2f1db722acbf4f7439a68ca752c2115a6c2869f249aa","abstract_canon_sha256":"f6bd062ef84920f800e7e828659bfe62af8819b38788cbce2b7d744258c9881b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:37.925504Z","signature_b64":"E3Uo035iOI12fRxFMBu1J8OGhxHSVf3C2YS0diGF0Y7uUUzrO+ayDGkAarfLYVJ3qwSEV/CZcsvAwQTsXr5fAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ef370b3dc6f8fd2f83e678ef352be1a56bff1f451bc52f702f9e0d6429714746","last_reissued_at":"2026-07-05T11:20:37.924980Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:37.924980Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Breaking Distortion-free Watermarks in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Dung Daniel T. Ngo, Guang Cheng, Hengzhi He, Manuela Veloso, Niccol\\`o Dalmasso, Saheed Obitayo, Shayleen Reynolds, Vamsi K. Potluru","submitted_at":"2025-02-25T19:52:55Z","abstract_excerpt":"In recent years, LLM watermarking has emerged as an attractive safeguard against AI-generated content, with promising applications in many real-world domains. However, there are growing concerns that the current LLM watermarking schemes are vulnerable to expert adversaries wishing to reverse-engineer the watermarking mechanisms. Prior work in breaking or stealing LLM watermarks mainly focuses on the distribution-modifying algorithm of Kirchenbauer et al. (2023), which perturbs the logit vector before sampling. In this work, we focus on reverse-engineering the other prominent LLM watermarking s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.18608","kind":"arxiv","version":2},"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/2502.18608/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":"2502.18608","created_at":"2026-07-05T11:20:37.925040+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.18608v2","created_at":"2026-07-05T11:20:37.925040+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.18608","created_at":"2026-07-05T11:20:37.925040+00:00"},{"alias_kind":"pith_short_12","alias_value":"543QWPOG7D6S","created_at":"2026-07-05T11:20:37.925040+00:00"},{"alias_kind":"pith_short_16","alias_value":"543QWPOG7D6S7A7G","created_at":"2026-07-05T11:20:37.925040+00:00"},{"alias_kind":"pith_short_8","alias_value":"543QWPOG","created_at":"2026-07-05T11:20:37.925040+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18430","citing_title":"Signature filtering: a lightweight enhancement for statistical watermark detection in large language models","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV","json":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV.json","graph_json":"https://pith.science/api/pith-number/543QWPOG7D6S7A7GPDXTKK7BUV/graph.json","events_json":"https://pith.science/api/pith-number/543QWPOG7D6S7A7GPDXTKK7BUV/events.json","paper":"https://pith.science/paper/543QWPOG"},"agent_actions":{"view_html":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV","download_json":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV.json","view_paper":"https://pith.science/paper/543QWPOG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.18608&json=true","fetch_graph":"https://pith.science/api/pith-number/543QWPOG7D6S7A7GPDXTKK7BUV/graph.json","fetch_events":"https://pith.science/api/pith-number/543QWPOG7D6S7A7GPDXTKK7BUV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV/action/storage_attestation","attest_author":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV/action/author_attestation","sign_citation":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV/action/citation_signature","submit_replication":"https://pith.science/pith/543QWPOG7D6S7A7GPDXTKK7BUV/action/replication_record"}},"created_at":"2026-07-05T11:20:37.925040+00:00","updated_at":"2026-07-05T11:20:37.925040+00:00"}