{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:CXFTK4TLNZ7F65OGRA2ZENKB2X","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"4da9a12fa614291c90dfc9cc3d1708c6affb25617f17f5b59d811a1e74a74abc","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-07T17:18:48Z","title_canon_sha256":"bebb938a02d07e22ceb13799065bb32a158da04327390e7919ffbe8c73847658"},"schema_version":"1.0","source":{"id":"2505.04578","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.04578","created_at":"2026-07-05T10:59:54Z"},{"alias_kind":"arxiv_version","alias_value":"2505.04578v1","created_at":"2026-07-05T10:59:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.04578","created_at":"2026-07-05T10:59:54Z"},{"alias_kind":"pith_short_12","alias_value":"CXFTK4TLNZ7F","created_at":"2026-07-05T10:59:54Z"},{"alias_kind":"pith_short_16","alias_value":"CXFTK4TLNZ7F65OG","created_at":"2026-07-05T10:59:54Z"},{"alias_kind":"pith_short_8","alias_value":"CXFTK4TL","created_at":"2026-07-05T10:59:54Z"}],"graph_snapshots":[{"event_id":"sha256:359b62e2cc245d60b894e88441df76709b57947f911b045f72a4fdc34e0b703f","target":"graph","created_at":"2026-07-05T10:59:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2505.04578/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement learning (RL) fine-tuning transforms large language models while creating a vulnerability we experimentally verify: Our experiment shows that malicious RL fine-tuning dismantles safety guardrails with remarkable efficiency, requiring only 50 steps and minimal adversarial prompts, with harmful escalating from 0-2 to 7-9. This attack vector particularly threatens open-source models with parameter-level access. Existing defenses targeting supervised fine-tuning prove ineffective against RL's dynamic feedback mechanisms. We introduce Reward Neutralization, the first defense framework","authors_text":"Wenjun Cao","cross_cats":["cs.AI"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-07T17:18:48Z","title":"Fight Fire with Fire: Defending Against Malicious RL Fine-Tuning via Reward Neutralization"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.04578","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:1487554ce47179af3455ed83aec605dc8194fa525e5d07efa17452f239827f26","target":"record","created_at":"2026-07-05T10:59:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"4da9a12fa614291c90dfc9cc3d1708c6affb25617f17f5b59d811a1e74a74abc","cross_cats_sorted":["cs.AI"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-07T17:18:48Z","title_canon_sha256":"bebb938a02d07e22ceb13799065bb32a158da04327390e7919ffbe8c73847658"},"schema_version":"1.0","source":{"id":"2505.04578","kind":"arxiv","version":1}},"canonical_sha256":"15cb35726b6e7e5f75c68835923541d5f6d1ae0bad5344287b7f56adfd02c7c8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"15cb35726b6e7e5f75c68835923541d5f6d1ae0bad5344287b7f56adfd02c7c8","first_computed_at":"2026-07-05T10:59:54.114565Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:59:54.114565Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"3jJBgtfwi8MoNjSk9iMewXmlM3uZ5kCEXNcYs87BvvBKx0cbff3aAbSoD6sXomXIIePO/MMyovEAIBf5EpvgDQ==","signature_status":"signed_v1","signed_at":"2026-07-05T10:59:54.115004Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.04578","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1487554ce47179af3455ed83aec605dc8194fa525e5d07efa17452f239827f26","sha256:359b62e2cc245d60b894e88441df76709b57947f911b045f72a4fdc34e0b703f"],"state_sha256":"0499d016d46ad13100337676553553ee60f697437d199eb5ccc75e1a6a57326c"}