{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:OGLUKIAHTLDA3WW5KB7BF7HS3D","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":"e3c079ae2600bf275b08d69cd35186b06c6e0866a8fc8437aeb3883148e4eabc","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-08T03:42:55Z","title_canon_sha256":"a10f9af0509997dc293432f4c36a1b1fab03222629d0056c181857d1aac8d9b0"},"schema_version":"1.0","source":{"id":"2603.07445","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.07445","created_at":"2026-06-04T01:09:48Z"},{"alias_kind":"arxiv_version","alias_value":"2603.07445v2","created_at":"2026-06-04T01:09:48Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.07445","created_at":"2026-06-04T01:09:48Z"},{"alias_kind":"pith_short_12","alias_value":"OGLUKIAHTLDA","created_at":"2026-06-04T01:09:48Z"},{"alias_kind":"pith_short_16","alias_value":"OGLUKIAHTLDA3WW5","created_at":"2026-06-04T01:09:48Z"},{"alias_kind":"pith_short_8","alias_value":"OGLUKIAH","created_at":"2026-06-04T01:09:48Z"}],"graph_snapshots":[{"event_id":"sha256:5b313eb502d5237849c4f3abf20e30c3a4243edd2b4c806e79310ade7267be87","target":"graph","created_at":"2026-06-04T01:09:48Z","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/2603.07445/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large language models (LLMs) often require fine-tuning (FT) to perform well on downstream tasks, but FT can induce safety-alignment drift even when the training dataset contains only benign data. Prior work shows that introducing a small fraction of harmful data can substantially compromise LLM refusal behavior, causing LLMs to comply with harmful requests. Existing defense methods often rely on model-wide interventions, such as restricting which parameters are updated or injecting additional safety data, which can limit generality and degrade downstream task performance. To address these limi","authors_text":"An Wang, Guoli Wang, Haonan Shi, Tu Ouyang","cross_cats":["cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-08T03:42:55Z","title":"Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.07445","kind":"arxiv","version":2},"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":"sha256:341bf13e8d1c5e009efa2d23fa6ab1b9549e962396b86ac697125f840a1c2bde","target":"record","created_at":"2026-06-04T01:09:48Z","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":"e3c079ae2600bf275b08d69cd35186b06c6e0866a8fc8437aeb3883148e4eabc","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-08T03:42:55Z","title_canon_sha256":"a10f9af0509997dc293432f4c36a1b1fab03222629d0056c181857d1aac8d9b0"},"schema_version":"1.0","source":{"id":"2603.07445","kind":"arxiv","version":2}},"canonical_sha256":"71974520079ac60ddadd507e12fcf2d8f0a0e9d09ff379257ef8c32e25004dc1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"71974520079ac60ddadd507e12fcf2d8f0a0e9d09ff379257ef8c32e25004dc1","first_computed_at":"2026-06-04T01:09:48.022479Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-04T01:09:48.022479Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"YZRckPpU3mI+3K724pZMx6IiJ8q3HWEkEUO6qvdqTSqm4MwBDR2tX2vrm6XQipX8fpPVuX4+SyfjUhiXOBViAg==","signature_status":"signed_v1","signed_at":"2026-06-04T01:09:48.023701Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.07445","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:341bf13e8d1c5e009efa2d23fa6ab1b9549e962396b86ac697125f840a1c2bde","sha256:5b313eb502d5237849c4f3abf20e30c3a4243edd2b4c806e79310ade7267be87"],"state_sha256":"aee1b0eb03da862eaf17e7fd3f923ed928ec6687b5512a6bfe4f44b7691f164b"}