{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YX4JGPDDGLO2SOGGBPREXGOOMA","short_pith_number":"pith:YX4JGPDD","schema_version":"1.0","canonical_sha256":"c5f8933c6332dda938c60be24b99ce601d5a88ac26abcbcb3174237895389916","source":{"kind":"arxiv","id":"2601.22313","version":2},"attestation_state":"computed","paper":{"title":"Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Duygu Nur Yaldiz, Eleni Triantafillou, Peter Kairouz, Sai Praneeth Karimireddy, Salman Avestimehr, Yavuz Bakman","submitted_at":"2026-01-29T20:54:21Z","abstract_excerpt":"Large Language Models (LLMs) are rarely static and are frequently updated in practice. A growing body of alignment research has shown that models initially deemed ``aligned'' can exhibit misaligned behavior after fine-tuning. These works typically assume that the initial model is aligned based on static black-box evaluation, i.e., the absence of undesired responses to a fixed set of queries. However, the limits of black-box evaluation for post-update scenarios is not explored sufficiently. In this work, we formalize model alignment in both the static and post-update settings and uncover a fund"},"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":"2601.22313","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-01-29T20:54:21Z","cross_cats_sorted":[],"title_canon_sha256":"4a87ef4cf56d333b96e8c66fac6b2ea5f0e424d084d394fb9fe22d6f0ab08996","abstract_canon_sha256":"3931e70d4ab8899465aff75dda7b99ebe8aef3e0e18975467f42024a0c436e67"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-13T00:17:07.238895Z","signature_b64":"DyX1GmcPLYxv6UohwGq24O3j3ZsmYdsEc4o3lUpGEkS/BlGlj+Wmouzw6mNe++fWAb2/YJTMb/LA/WvEzqiCDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c5f8933c6332dda938c60be24b99ce601d5a88ac26abcbcb3174237895389916","last_reissued_at":"2026-07-13T00:17:07.237041Z","signature_status":"signed_v1","first_computed_at":"2026-07-13T00:17:07.237041Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Duygu Nur Yaldiz, Eleni Triantafillou, Peter Kairouz, Sai Praneeth Karimireddy, Salman Avestimehr, Yavuz Bakman","submitted_at":"2026-01-29T20:54:21Z","abstract_excerpt":"Large Language Models (LLMs) are rarely static and are frequently updated in practice. A growing body of alignment research has shown that models initially deemed ``aligned'' can exhibit misaligned behavior after fine-tuning. These works typically assume that the initial model is aligned based on static black-box evaluation, i.e., the absence of undesired responses to a fixed set of queries. However, the limits of black-box evaluation for post-update scenarios is not explored sufficiently. In this work, we formalize model alignment in both the static and post-update settings and uncover a fund"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.22313","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":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.22313/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":"2601.22313","created_at":"2026-07-13T00:17:07.238093+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.22313v2","created_at":"2026-07-13T00:17:07.238093+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.22313","created_at":"2026-07-13T00:17:07.238093+00:00"},{"alias_kind":"pith_short_12","alias_value":"YX4JGPDDGLO2","created_at":"2026-07-13T00:17:07.238093+00:00"},{"alias_kind":"pith_short_16","alias_value":"YX4JGPDDGLO2SOGG","created_at":"2026-07-13T00:17:07.238093+00:00"},{"alias_kind":"pith_short_8","alias_value":"YX4JGPDD","created_at":"2026-07-13T00:17:07.238093+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.09866","citing_title":"Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning","ref_index":90,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA","json":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA.json","graph_json":"https://pith.science/api/pith-number/YX4JGPDDGLO2SOGGBPREXGOOMA/graph.json","events_json":"https://pith.science/api/pith-number/YX4JGPDDGLO2SOGGBPREXGOOMA/events.json","paper":"https://pith.science/paper/YX4JGPDD"},"agent_actions":{"view_html":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA","download_json":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA.json","view_paper":"https://pith.science/paper/YX4JGPDD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.22313&json=true","fetch_graph":"https://pith.science/api/pith-number/YX4JGPDDGLO2SOGGBPREXGOOMA/graph.json","fetch_events":"https://pith.science/api/pith-number/YX4JGPDDGLO2SOGGBPREXGOOMA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA/action/storage_attestation","attest_author":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA/action/author_attestation","sign_citation":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA/action/citation_signature","submit_replication":"https://pith.science/pith/YX4JGPDDGLO2SOGGBPREXGOOMA/action/replication_record"}},"created_at":"2026-07-13T00:17:07.238093+00:00","updated_at":"2026-07-13T00:17:07.238093+00:00"}