{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3IBYXXN32UY4ZBEZA7VJNRANUO","short_pith_number":"pith:3IBYXXN3","schema_version":"1.0","canonical_sha256":"da038bddbbd531cc849907ea96c40da38cbe31b809c61b12e1134e373868029c","source":{"kind":"arxiv","id":"2606.20814","version":1},"attestation_state":"computed","paper":{"title":"What Shapes Emergent Misalignment? Insights from Training Dynamics, Model Priors, and Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Anietta Weckauff, Diego Garcia-Olano, Maksym Andriushchenko, Yuchen Zhang","submitted_at":"2026-06-18T18:04:20Z","abstract_excerpt":"Emergent misalignment (EM) is a phenomenon in which models generalize with narrow fine-tuning, leading to broad (yet uneven) misalignment across evaluation questions. We study EM and its variability directly through the components of fine-tuning: training dynamics, model priors, and data. (1) We first explored how in-domain training loss relates to out-of-domain alignment scores across datasets and model families. Then, we tried to induce potential alternative local minima through different learning schedules for one narrow fine-tuning, but did not find strong runs with better broad alignment "},"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":"2606.20814","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-06-18T18:04:20Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"a0355d2f359548f87be6995c2e3208785fc9796e6b313d00bb007f4cfd24e9a4","abstract_canon_sha256":"5e1aeceadc48874acfaa50798daac3ce5563670b898e8a683791defdac19b7b6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T00:11:59.598089Z","signature_b64":"mS6WOzc+PflAcmdPKzLeX+YY/fir4C9cg0zgX8GrMVbPrkUhiIL86rdv04tJHOfuTaDVXVktpROS36eEIX7DCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da038bddbbd531cc849907ea96c40da38cbe31b809c61b12e1134e373868029c","last_reissued_at":"2026-06-23T00:11:59.597688Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T00:11:59.597688Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What Shapes Emergent Misalignment? Insights from Training Dynamics, Model Priors, and Data","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Anietta Weckauff, Diego Garcia-Olano, Maksym Andriushchenko, Yuchen Zhang","submitted_at":"2026-06-18T18:04:20Z","abstract_excerpt":"Emergent misalignment (EM) is a phenomenon in which models generalize with narrow fine-tuning, leading to broad (yet uneven) misalignment across evaluation questions. We study EM and its variability directly through the components of fine-tuning: training dynamics, model priors, and data. (1) We first explored how in-domain training loss relates to out-of-domain alignment scores across datasets and model families. Then, we tried to induce potential alternative local minima through different learning schedules for one narrow fine-tuning, but did not find strong runs with better broad alignment "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.20814","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/2606.20814/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":"2606.20814","created_at":"2026-06-23T00:11:59.597741+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.20814v1","created_at":"2026-06-23T00:11:59.597741+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.20814","created_at":"2026-06-23T00:11:59.597741+00:00"},{"alias_kind":"pith_short_12","alias_value":"3IBYXXN32UY4","created_at":"2026-06-23T00:11:59.597741+00:00"},{"alias_kind":"pith_short_16","alias_value":"3IBYXXN32UY4ZBEZ","created_at":"2026-06-23T00:11:59.597741+00:00"},{"alias_kind":"pith_short_8","alias_value":"3IBYXXN3","created_at":"2026-06-23T00:11:59.597741+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/3IBYXXN32UY4ZBEZA7VJNRANUO","json":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO.json","graph_json":"https://pith.science/api/pith-number/3IBYXXN32UY4ZBEZA7VJNRANUO/graph.json","events_json":"https://pith.science/api/pith-number/3IBYXXN32UY4ZBEZA7VJNRANUO/events.json","paper":"https://pith.science/paper/3IBYXXN3"},"agent_actions":{"view_html":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO","download_json":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO.json","view_paper":"https://pith.science/paper/3IBYXXN3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.20814&json=true","fetch_graph":"https://pith.science/api/pith-number/3IBYXXN32UY4ZBEZA7VJNRANUO/graph.json","fetch_events":"https://pith.science/api/pith-number/3IBYXXN32UY4ZBEZA7VJNRANUO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO/action/storage_attestation","attest_author":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO/action/author_attestation","sign_citation":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO/action/citation_signature","submit_replication":"https://pith.science/pith/3IBYXXN32UY4ZBEZA7VJNRANUO/action/replication_record"}},"created_at":"2026-06-23T00:11:59.597741+00:00","updated_at":"2026-06-23T00:11:59.597741+00:00"}