{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XZMSQQEMCETZRL3COIRVWQQQU7","short_pith_number":"pith:XZMSQQEM","schema_version":"1.0","canonical_sha256":"be5928408c112798af6272235b4210a7f7dc74f94ae69cce72f7cc16873716b9","source":{"kind":"arxiv","id":"2509.08000","version":1},"attestation_state":"computed","paper":{"title":"AntiDote: Bi-level Adversarial Training for Tamper-Resistant LLMs","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Debdeep Sanyal, Manodeep Ray, Murari Mandal","submitted_at":"2025-09-06T16:03:07Z","abstract_excerpt":"The release of open-weight large language models (LLMs) creates a tension between advancing accessible research and preventing misuse, such as malicious fine-tuning to elicit harmful content. Current safety measures struggle to preserve the general capabilities of the LLM while resisting a determined adversary with full access to the model's weights and architecture, who can use full-parameter fine-tuning to erase existing safeguards. To address this, we introduce AntiDote, a bi-level optimization procedure for training LLMs to be resistant to such tampering. AntiDote involves an auxiliary adv"},"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":"2509.08000","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-09-06T16:03:07Z","cross_cats_sorted":[],"title_canon_sha256":"72c838fced72d4c5b27b5981fc4445dfb6996df61ee43efe0a6ebcc2d204b3a3","abstract_canon_sha256":"a5dd87b86da0be0e102efd3a8443fa3a2799c68cedc3f349056097e69b46f33d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:08:09.476881Z","signature_b64":"X6RxV3lLhmcGBNtNpzTpuEVQ4M499k+6QltfXX86MoxxKcX7hCAQNWo/28rGhrEyjESlpPXlZNQgufxMj/nTBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"be5928408c112798af6272235b4210a7f7dc74f94ae69cce72f7cc16873716b9","last_reissued_at":"2026-07-05T12:08:09.476219Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:08:09.476219Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AntiDote: Bi-level Adversarial Training for Tamper-Resistant LLMs","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Debdeep Sanyal, Manodeep Ray, Murari Mandal","submitted_at":"2025-09-06T16:03:07Z","abstract_excerpt":"The release of open-weight large language models (LLMs) creates a tension between advancing accessible research and preventing misuse, such as malicious fine-tuning to elicit harmful content. Current safety measures struggle to preserve the general capabilities of the LLM while resisting a determined adversary with full access to the model's weights and architecture, who can use full-parameter fine-tuning to erase existing safeguards. To address this, we introduce AntiDote, a bi-level optimization procedure for training LLMs to be resistant to such tampering. AntiDote involves an auxiliary adv"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.08000","kind":"arxiv","version":1},"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/2509.08000/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":"2509.08000","created_at":"2026-07-05T12:08:09.476307+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.08000v1","created_at":"2026-07-05T12:08:09.476307+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.08000","created_at":"2026-07-05T12:08:09.476307+00:00"},{"alias_kind":"pith_short_12","alias_value":"XZMSQQEMCETZ","created_at":"2026-07-05T12:08:09.476307+00:00"},{"alias_kind":"pith_short_16","alias_value":"XZMSQQEMCETZRL3C","created_at":"2026-07-05T12:08:09.476307+00:00"},{"alias_kind":"pith_short_8","alias_value":"XZMSQQEM","created_at":"2026-07-05T12:08:09.476307+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.15980","citing_title":"Do Activation Monitors Survive Model Updates? Benchmarking, Predicting, and Repairing Activation-Monitor Staleness","ref_index":54,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7","json":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7.json","graph_json":"https://pith.science/api/pith-number/XZMSQQEMCETZRL3COIRVWQQQU7/graph.json","events_json":"https://pith.science/api/pith-number/XZMSQQEMCETZRL3COIRVWQQQU7/events.json","paper":"https://pith.science/paper/XZMSQQEM"},"agent_actions":{"view_html":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7","download_json":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7.json","view_paper":"https://pith.science/paper/XZMSQQEM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.08000&json=true","fetch_graph":"https://pith.science/api/pith-number/XZMSQQEMCETZRL3COIRVWQQQU7/graph.json","fetch_events":"https://pith.science/api/pith-number/XZMSQQEMCETZRL3COIRVWQQQU7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7/action/storage_attestation","attest_author":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7/action/author_attestation","sign_citation":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7/action/citation_signature","submit_replication":"https://pith.science/pith/XZMSQQEMCETZRL3COIRVWQQQU7/action/replication_record"}},"created_at":"2026-07-05T12:08:09.476307+00:00","updated_at":"2026-07-05T12:08:09.476307+00:00"}