{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:U74HG3VYFTMKGXJEAURRDPQOTL","short_pith_number":"pith:U74HG3VY","schema_version":"1.0","canonical_sha256":"a7f8736eb82cd8a35d24052311be0e9af08e8bea852d92bb86f3baaa94f2f5a1","source":{"kind":"arxiv","id":"2509.20924","version":2},"attestation_state":"computed","paper":{"title":"RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hanbo Huang, Hao Zheng, Lin Liu, Shiyu Liang, Xuan Gong, Yihan Li, Yiran Zhang, Zhuotao Liu","submitted_at":"2025-09-25T09:08:02Z","abstract_excerpt":"Large language model (LLM) watermarking has shown promise in detecting AI-generated content and mitigating misuse, with prior work claiming robustness against paraphrasing and text editing. In this paper, we argue that existing evaluations are not sufficiently adversarial, obscuring critical vulnerabilities and overstating the security. To address this, we introduce the adaptive robustness radius, a formal metric that quantifies the worst-case resilience of watermarks against adaptive adversaries. By lifting the paraphrase space into a KL-divergence ball, we approximate this radius and theoret"},"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":"2509.20924","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CR","submitted_at":"2025-09-25T09:08:02Z","cross_cats_sorted":[],"title_canon_sha256":"2f2cb23753efc5f94770fa5362e696c215247eb143a4345b35a09f6c6950f134","abstract_canon_sha256":"69a5861bdc1346afd4ee38004f83422fbf816aba6a1c2d4ad15bffb542ab1602"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:17.441452Z","signature_b64":"80egN/qTNex6aRv41bIAe+xxjZ5Ba5g+gZ6iWixJJ5u1l827NmGaY4GeDBM6LRrr70qwkIupytWLS9H7YFBdDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7f8736eb82cd8a35d24052311be0e9af08e8bea852d92bb86f3baaa94f2f5a1","last_reissued_at":"2026-05-17T23:39:17.440906Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:17.440906Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RLCracker: Evaluating the Worst-Case Vulnerability of LLM Watermarks with Adaptive RL Attacks","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Hanbo Huang, Hao Zheng, Lin Liu, Shiyu Liang, Xuan Gong, Yihan Li, Yiran Zhang, Zhuotao Liu","submitted_at":"2025-09-25T09:08:02Z","abstract_excerpt":"Large language model (LLM) watermarking has shown promise in detecting AI-generated content and mitigating misuse, with prior work claiming robustness against paraphrasing and text editing. In this paper, we argue that existing evaluations are not sufficiently adversarial, obscuring critical vulnerabilities and overstating the security. To address this, we introduce the adaptive robustness radius, a formal metric that quantifies the worst-case resilience of watermarks against adaptive adversaries. By lifting the paraphrase space into a KL-divergence ball, we approximate this radius and theoret"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.20924","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":""},"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":"2509.20924","created_at":"2026-05-17T23:39:17.440980+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.20924v2","created_at":"2026-05-17T23:39:17.440980+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.20924","created_at":"2026-05-17T23:39:17.440980+00:00"},{"alias_kind":"pith_short_12","alias_value":"U74HG3VYFTMK","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"U74HG3VYFTMKGXJE","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"U74HG3VY","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2604.11546","citing_title":"RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience","ref_index":17,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL","json":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL.json","graph_json":"https://pith.science/api/pith-number/U74HG3VYFTMKGXJEAURRDPQOTL/graph.json","events_json":"https://pith.science/api/pith-number/U74HG3VYFTMKGXJEAURRDPQOTL/events.json","paper":"https://pith.science/paper/U74HG3VY"},"agent_actions":{"view_html":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL","download_json":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL.json","view_paper":"https://pith.science/paper/U74HG3VY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.20924&json=true","fetch_graph":"https://pith.science/api/pith-number/U74HG3VYFTMKGXJEAURRDPQOTL/graph.json","fetch_events":"https://pith.science/api/pith-number/U74HG3VYFTMKGXJEAURRDPQOTL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL/action/storage_attestation","attest_author":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL/action/author_attestation","sign_citation":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL/action/citation_signature","submit_replication":"https://pith.science/pith/U74HG3VYFTMKGXJEAURRDPQOTL/action/replication_record"}},"created_at":"2026-05-17T23:39:17.440980+00:00","updated_at":"2026-05-17T23:39:17.440980+00:00"}