{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:ERNFVBFKEPMLFXMRQUEW3VDMX5","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":"2e0aed015bf1a11ff1a2707dd39ac6cec235d990c6266eb21325da8d7d016f11","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-03T23:16:14Z","title_canon_sha256":"1932e8f8bef3c64a24772dfe65d8518d6aa60796de036002fd399059937a2d6e"},"schema_version":"1.0","source":{"id":"2605.02087","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.02087","created_at":"2026-05-25T02:01:21Z"},{"alias_kind":"arxiv_version","alias_value":"2605.02087v2","created_at":"2026-05-25T02:01:21Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.02087","created_at":"2026-05-25T02:01:21Z"},{"alias_kind":"pith_short_12","alias_value":"ERNFVBFKEPML","created_at":"2026-05-25T02:01:21Z"},{"alias_kind":"pith_short_16","alias_value":"ERNFVBFKEPMLFXMR","created_at":"2026-05-25T02:01:21Z"},{"alias_kind":"pith_short_8","alias_value":"ERNFVBFK","created_at":"2026-05-25T02:01:21Z"}],"graph_snapshots":[{"event_id":"sha256:81bc5fd3c6ffb747637859668e1afdf9906ae6b7672b7a7452c4f2d4cdfd10e5","target":"graph","created_at":"2026-05-25T02:01:21Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"applying MSM with a spec addressing self-preservation and goal-guarding substantially reduces agentic misalignment rate (Qwen3-32B: 54% to 7%), beating a deliberative alignment baseline (14%)."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That training on synthetic documents discussing the Model Spec will reliably encode the intended generalizations into the model without introducing new unintended behaviors or degrading other capabilities."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Model spec midtraining trains AI models on documents about their alignment rules before demonstration fine-tuning, producing stronger and more controllable generalization to the intended values and safety behaviors."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples."}],"snapshot_sha256":"7f628945b51e438d630ee4d9120457419a2a2519e440f7c4a3725ecadeb3277d"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cd9373de7701a1add7448ff08ff170df1bab33080f297e6571a9ffa56834bff4"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T16:38:00.882396Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-20T04:01:22.783057Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T16:40:46.733574Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.02087/integrity.json","findings":[],"snapshot_sha256":"c9036558d6cb50e7cd8492ddc1f6f4422dc4bfc4f2c3f3b9aa044e511877a68d","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Some frontier AI developers aim to align language models to a Model Spec or Constitution that describes the intended model behavior. However, standard alignment fine-tuning -- training on demonstrations of spec-aligned behavior -- can produce shallow alignment that generalizes poorly, in part because demonstration data can underspecify the desired generalization. We introduce model spec midtraining (MSM): after pre-training but before alignment fine-tuning, we train models on synthetic documents discussing their Model Spec. This teaches models the content of the spec, thereby shaping how they ","authors_text":"Chloe Li, Jon Kutasov, Nevan Wichers, Samuel Marks, Sara Price","cross_cats":[],"headline":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-03T23:16:14Z","title":"Model Spec Midtraining: Improving How Alignment Training Generalizes"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.02087","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-08T19:10:59.393102Z","id":"dc57bf1f-cdbd-45c8-bb9f-eb5e43c3efda","model_set":{"reader":"grok-4.3"},"one_line_summary":"Model spec midtraining trains AI models on documents about their alignment rules before demonstration fine-tuning, producing stronger and more controllable generalization to the intended values and safety behaviors.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples.","strongest_claim":"applying MSM with a spec addressing self-preservation and goal-guarding substantially reduces agentic misalignment rate (Qwen3-32B: 54% to 7%), beating a deliberative alignment baseline (14%).","weakest_assumption":"That training on synthetic documents discussing the Model Spec will reliably encode the intended generalizations into the model without introducing new unintended behaviors or degrading other capabilities."}},"verdict_id":"dc57bf1f-cdbd-45c8-bb9f-eb5e43c3efda"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:e683111eab5a929e319c0ec55fb73e8fdc556f7195bb966e8ae52bc9ad62518a","target":"record","created_at":"2026-05-25T02:01:21Z","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":"2e0aed015bf1a11ff1a2707dd39ac6cec235d990c6266eb21325da8d7d016f11","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-03T23:16:14Z","title_canon_sha256":"1932e8f8bef3c64a24772dfe65d8518d6aa60796de036002fd399059937a2d6e"},"schema_version":"1.0","source":{"id":"2605.02087","kind":"arxiv","version":2}},"canonical_sha256":"245a5a84aa23d8b2dd9185096dd46cbf4cfff3a210b0efb857f6bce56b92d806","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"245a5a84aa23d8b2dd9185096dd46cbf4cfff3a210b0efb857f6bce56b92d806","first_computed_at":"2026-05-25T02:01:21.550661Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-25T02:01:21.550661Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"g5Q+0VhCXEbZLaIhEIixKjw2YPcxqG8TyHuJf+WpMfsL0AqkAgUO++OrxS5tI/feNLeGA9lf+A3CZtNkEvy+Dg==","signature_status":"signed_v1","signed_at":"2026-05-25T02:01:21.551452Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.02087","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:e683111eab5a929e319c0ec55fb73e8fdc556f7195bb966e8ae52bc9ad62518a","sha256:81bc5fd3c6ffb747637859668e1afdf9906ae6b7672b7a7452c4f2d4cdfd10e5"],"state_sha256":"efb55d59ebc33eb822ab4d326aeaaa26c3601a820390cec9bc2da59aafad8b39"}