{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:ERNFVBFKEPMLFXMRQUEW3VDMX5","short_pith_number":"pith:ERNFVBFK","canonical_record":{"source":{"id":"2605.02087","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-03T23:16:14Z","cross_cats_sorted":[],"title_canon_sha256":"1932e8f8bef3c64a24772dfe65d8518d6aa60796de036002fd399059937a2d6e","abstract_canon_sha256":"2e0aed015bf1a11ff1a2707dd39ac6cec235d990c6266eb21325da8d7d016f11"},"schema_version":"1.0"},"canonical_sha256":"245a5a84aa23d8b2dd9185096dd46cbf4cfff3a210b0efb857f6bce56b92d806","source":{"kind":"arxiv","id":"2605.02087","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"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:ERNFVBFKEPMLFXMRQUEW3VDMX5","target":"record","payload":{"canonical_record":{"source":{"id":"2605.02087","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-03T23:16:14Z","cross_cats_sorted":[],"title_canon_sha256":"1932e8f8bef3c64a24772dfe65d8518d6aa60796de036002fd399059937a2d6e","abstract_canon_sha256":"2e0aed015bf1a11ff1a2707dd39ac6cec235d990c6266eb21325da8d7d016f11"},"schema_version":"1.0"},"canonical_sha256":"245a5a84aa23d8b2dd9185096dd46cbf4cfff3a210b0efb857f6bce56b92d806","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:21.551452Z","signature_b64":"g5Q+0VhCXEbZLaIhEIixKjw2YPcxqG8TyHuJf+WpMfsL0AqkAgUO++OrxS5tI/feNLeGA9lf+A3CZtNkEvy+Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"245a5a84aa23d8b2dd9185096dd46cbf4cfff3a210b0efb857f6bce56b92d806","last_reissued_at":"2026-05-25T02:01:21.550661Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:21.550661Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.02087","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-25T02:01:21Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"DE6kPJS4DlMYIQb1Pkba4zG0+O/MesdRONwJvgR+4rIKrndzoXr9L+ZeoV2vLIve4quoewOhAcH3SB/yChePAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-29T10:50:51.403950Z"},"content_sha256":"e683111eab5a929e319c0ec55fb73e8fdc556f7195bb966e8ae52bc9ad62518a","schema_version":"1.0","event_id":"sha256:e683111eab5a929e319c0ec55fb73e8fdc556f7195bb966e8ae52bc9ad62518a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:ERNFVBFKEPMLFXMRQUEW3VDMX5","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Model Spec Midtraining: Improving How Alignment Training Generalizes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chloe Li, Jon Kutasov, Nevan Wichers, Samuel Marks, Sara Price","submitted_at":"2026-05-03T23:16:14Z","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 "},"claims":{"count":4,"items":[{"kind":"strongest_claim","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%).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","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.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","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.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7f628945b51e438d630ee4d9120457419a2a2519e440f7c4a3725ecadeb3277d"},"source":{"id":"2605.02087","kind":"arxiv","version":2},"verdict":{"id":"dc57bf1f-cdbd-45c8-bb9f-eb5e43c3efda","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T19:10:59.393102Z","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%).","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","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.","pith_extraction_headline":"Training models on synthetic documents about their Model Spec before alignment fine-tuning shapes how they generalize from later examples."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.02087/integrity.json","findings":[],"available":true,"detectors_run":[{"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","findings_count":0}],"snapshot_sha256":"c9036558d6cb50e7cd8492ddc1f6f4422dc4bfc4f2c3f3b9aa044e511877a68d"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cd9373de7701a1add7448ff08ff170df1bab33080f297e6571a9ffa56834bff4"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"dc57bf1f-cdbd-45c8-bb9f-eb5e43c3efda"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-25T02:01:21Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Erg5sGupVtDCEUeHLwB5vmzjKh+i696j3+jXGfSBuC7ZRg1rcnu+8Ob3cGqvlRWv4dwDYosAxUkq21aJFEXHBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-29T10:50:51.404431Z"},"content_sha256":"81bc5fd3c6ffb747637859668e1afdf9906ae6b7672b7a7452c4f2d4cdfd10e5","schema_version":"1.0","event_id":"sha256:81bc5fd3c6ffb747637859668e1afdf9906ae6b7672b7a7452c4f2d4cdfd10e5"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/bundle.json","state_url":"https://pith.science/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-29T10:50:51Z","links":{"resolver":"https://pith.science/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5","bundle":"https://pith.science/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/bundle.json","state":"https://pith.science/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/state.json","well_known_bundle":"https://pith.science/.well-known/pith/ERNFVBFKEPMLFXMRQUEW3VDMX5/bundle.json"},"state":{"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"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"80kET7MP79DSYIRu7WrtulhDC4hjnq/msbcAO/xRLFO0qBe07j8oMYD0OTMN1MK/oELVbJjaZ1z5g/U6yZuqBQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-29T10:50:51.406553Z","bundle_sha256":"5feb436f73c7946dc6f2b48f5f2feac5bdd11b0970a1c57e4019d2ceecef4945"}}