{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YEO77KHK7MTQKYIZ4LEF4RNCGV","short_pith_number":"pith:YEO77KHK","schema_version":"1.0","canonical_sha256":"c11dffa8eafb27056119e2c85e45a2357b53e1f6a5139858630a00376bb1cb33","source":{"kind":"arxiv","id":"2608.08542","version":1},"attestation_state":"computed","paper":{"title":"When Skills Meet Safety: Benchmarking and Characterizing the Adaptive Jailbreak Robustness of Skill-Merged LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongli Shi, Jing Li, Weiwei Hou, Xinran Xu, Yu Ma","submitted_at":"2026-08-09T07:41:06Z","abstract_excerpt":"Model merging has become the default way to give an aligned language model new skills without retraining: a practitioner folds task vectors from math, code, or domain specialists into a safety-aligned base using task arithmetic, TIES, or DARE. This convenience is known to carry a safety cost, but almost all of that evidence rests on static refusal tests: fixed harmful prompts scored for compliance. We argue this is misleading. Because safety alignment is \"shallow,\" concentrated in the first few generated tokens, a merged model's static refusal can stay clean while a real adaptive attack still "},"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":"2608.08542","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-08-09T07:41:06Z","cross_cats_sorted":[],"title_canon_sha256":"5da51cf5a11f030e3e7df75f8d36284c2b4456d690a6023757848309783f8463","abstract_canon_sha256":"0ddd85e971ce777d15f61decd702ada3fbdf65286fd7a8fcbe5db3d2c6ed6b10"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-11T01:22:55.363796Z","signature_b64":"bKiawt7VtPfAoxchydFd/Vl5QNfvmsKpxWH8/tPXu5HdknJYpXF7uZ+YpS+XNAD6VVyjARBP7zDDDL/qOkR/BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c11dffa8eafb27056119e2c85e45a2357b53e1f6a5139858630a00376bb1cb33","last_reissued_at":"2026-08-11T01:22:55.361347Z","signature_status":"signed_v1","first_computed_at":"2026-08-11T01:22:55.361347Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Skills Meet Safety: Benchmarking and Characterizing the Adaptive Jailbreak Robustness of Skill-Merged LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hongli Shi, Jing Li, Weiwei Hou, Xinran Xu, Yu Ma","submitted_at":"2026-08-09T07:41:06Z","abstract_excerpt":"Model merging has become the default way to give an aligned language model new skills without retraining: a practitioner folds task vectors from math, code, or domain specialists into a safety-aligned base using task arithmetic, TIES, or DARE. This convenience is known to carry a safety cost, but almost all of that evidence rests on static refusal tests: fixed harmful prompts scored for compliance. We argue this is misleading. Because safety alignment is \"shallow,\" concentrated in the first few generated tokens, a merged model's static refusal can stay clean while a real adaptive attack still "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.08542","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/2608.08542/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":"2608.08542","created_at":"2026-08-11T01:22:55.362105+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.08542v1","created_at":"2026-08-11T01:22:55.362105+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.08542","created_at":"2026-08-11T01:22:55.362105+00:00"},{"alias_kind":"pith_short_12","alias_value":"YEO77KHK7MTQ","created_at":"2026-08-11T01:22:55.362105+00:00"},{"alias_kind":"pith_short_16","alias_value":"YEO77KHK7MTQKYIZ","created_at":"2026-08-11T01:22:55.362105+00:00"},{"alias_kind":"pith_short_8","alias_value":"YEO77KHK","created_at":"2026-08-11T01:22:55.362105+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV","json":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV.json","graph_json":"https://pith.science/api/pith-number/YEO77KHK7MTQKYIZ4LEF4RNCGV/graph.json","events_json":"https://pith.science/api/pith-number/YEO77KHK7MTQKYIZ4LEF4RNCGV/events.json","paper":"https://pith.science/paper/YEO77KHK"},"agent_actions":{"view_html":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV","download_json":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV.json","view_paper":"https://pith.science/paper/YEO77KHK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.08542&json=true","fetch_graph":"https://pith.science/api/pith-number/YEO77KHK7MTQKYIZ4LEF4RNCGV/graph.json","fetch_events":"https://pith.science/api/pith-number/YEO77KHK7MTQKYIZ4LEF4RNCGV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV/action/storage_attestation","attest_author":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV/action/author_attestation","sign_citation":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV/action/citation_signature","submit_replication":"https://pith.science/pith/YEO77KHK7MTQKYIZ4LEF4RNCGV/action/replication_record"}},"created_at":"2026-08-11T01:22:55.362105+00:00","updated_at":"2026-08-11T01:22:55.362105+00:00"}