{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TI6CMAIFHFL25CLZ4RJA4WQIG6","short_pith_number":"pith:TI6CMAIF","schema_version":"1.0","canonical_sha256":"9a3c2601053957ae8979e4520e5a0837b1bee3fde6b8fb6ec84e04d54a4073ce","source":{"kind":"arxiv","id":"2508.20015","version":1},"attestation_state":"computed","paper":{"title":"Decomposing Behavioral Phase Transitions in LLMs: Order Parameters for Emergent Misalignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Julian Arnold, Niels L\\\"orch","submitted_at":"2025-08-27T16:19:49Z","abstract_excerpt":"Fine-tuning LLMs on narrowly harmful datasets can lead to behavior that is broadly misaligned with respect to human values. To understand when and how this emergent misalignment occurs, we develop a comprehensive framework for detecting and characterizing rapid transitions during fine-tuning using both distributional change detection methods as well as order parameters that are formulated in plain English and evaluated by an LLM judge. Using an objective statistical dissimilarity measure, we quantify how the phase transition that occurs during fine-tuning affects multiple aspects of the model."},"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":"2508.20015","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-08-27T16:19:49Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b51bea1f6bea9f0fe8ea608bbc675bd5fc78d3193e48f40510d5f55ff2aa53fa","abstract_canon_sha256":"ee1448f7caf16347446c7c4a7ab1a7a8518ad9d909cc7c51363a0a56c16182bf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:00:22.172162Z","signature_b64":"blqkmKG76DzjtnYhk0SdP3bJAqmopuVjcOf5W3M/9rgrHEb0qaeS+DdO5UsGlnzjpSEWVfuqZqvzM75Oc7S4BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a3c2601053957ae8979e4520e5a0837b1bee3fde6b8fb6ec84e04d54a4073ce","last_reissued_at":"2026-07-05T12:00:22.171633Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:00:22.171633Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Decomposing Behavioral Phase Transitions in LLMs: Order Parameters for Emergent Misalignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Julian Arnold, Niels L\\\"orch","submitted_at":"2025-08-27T16:19:49Z","abstract_excerpt":"Fine-tuning LLMs on narrowly harmful datasets can lead to behavior that is broadly misaligned with respect to human values. To understand when and how this emergent misalignment occurs, we develop a comprehensive framework for detecting and characterizing rapid transitions during fine-tuning using both distributional change detection methods as well as order parameters that are formulated in plain English and evaluated by an LLM judge. Using an objective statistical dissimilarity measure, we quantify how the phase transition that occurs during fine-tuning affects multiple aspects of the model."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.20015","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/2508.20015/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":"2508.20015","created_at":"2026-07-05T12:00:22.171694+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.20015v1","created_at":"2026-07-05T12:00:22.171694+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.20015","created_at":"2026-07-05T12:00:22.171694+00:00"},{"alias_kind":"pith_short_12","alias_value":"TI6CMAIFHFL2","created_at":"2026-07-05T12:00:22.171694+00:00"},{"alias_kind":"pith_short_16","alias_value":"TI6CMAIFHFL25CLZ","created_at":"2026-07-05T12:00:22.171694+00:00"},{"alias_kind":"pith_short_8","alias_value":"TI6CMAIF","created_at":"2026-07-05T12:00:22.171694+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09068","citing_title":"Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16325","citing_title":"Phase Transitions in Driven Informational Systems: A Two-Field Perspective on Learning Theory and Non-Equilibrium Chemistry","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12199","citing_title":"Overtrained, Not Misaligned","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6","json":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6.json","graph_json":"https://pith.science/api/pith-number/TI6CMAIFHFL25CLZ4RJA4WQIG6/graph.json","events_json":"https://pith.science/api/pith-number/TI6CMAIFHFL25CLZ4RJA4WQIG6/events.json","paper":"https://pith.science/paper/TI6CMAIF"},"agent_actions":{"view_html":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6","download_json":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6.json","view_paper":"https://pith.science/paper/TI6CMAIF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.20015&json=true","fetch_graph":"https://pith.science/api/pith-number/TI6CMAIFHFL25CLZ4RJA4WQIG6/graph.json","fetch_events":"https://pith.science/api/pith-number/TI6CMAIFHFL25CLZ4RJA4WQIG6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6/action/storage_attestation","attest_author":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6/action/author_attestation","sign_citation":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6/action/citation_signature","submit_replication":"https://pith.science/pith/TI6CMAIFHFL25CLZ4RJA4WQIG6/action/replication_record"}},"created_at":"2026-07-05T12:00:22.171694+00:00","updated_at":"2026-07-05T12:00:22.171694+00:00"}