{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:L33RNGUYBWS576LT36D4GTW3WL","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":"98133f29ab394c6fc1eb1e7453dce3e825395a16bc5f801aae67f197ba48a623","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-01T18:00:55Z","title_canon_sha256":"bbeb7847e2c0218b82d28e816a5d6a57d3dc09a8f6399c5e4101c304f0ba6784"},"schema_version":"1.0","source":{"id":"2605.00994","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.00994","created_at":"2026-06-30T02:18:08Z"},{"alias_kind":"arxiv_version","alias_value":"2605.00994v2","created_at":"2026-06-30T02:18:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.00994","created_at":"2026-06-30T02:18:08Z"},{"alias_kind":"pith_short_12","alias_value":"L33RNGUYBWS5","created_at":"2026-06-30T02:18:08Z"},{"alias_kind":"pith_short_16","alias_value":"L33RNGUYBWS576LT","created_at":"2026-06-30T02:18:08Z"},{"alias_kind":"pith_short_8","alias_value":"L33RNGUY","created_at":"2026-06-30T02:18:08Z"}],"graph_snapshots":[{"event_id":"sha256:6f1ed3a768707114d359baae71bf06eec6016af20ad31b8aa918afe1b1890e05","target":"graph","created_at":"2026-06-30T02:18:08Z","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":"a simple perplexity-based method can surface finetuning objectives from model organisms by leveraging their tendency to overgeneralize their finetuned behaviors beyond the intended context"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"that finetuned models reliably overgeneralize their training objectives to unrelated contexts generated from random prefills, producing detectable perplexity gaps against a reference model"},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Perplexity gaps between finetuned and reference models on random-prefill completions often reveal the original finetuning objectives across diverse model organisms."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms."}],"snapshot_sha256":"92ed26ec09b935a7177b025b6539c412aabe204323ca7a3a900a9dbf28128d73"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T18:40:27.363128Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T17:45:49.779381Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.00994/integrity.json","findings":[],"snapshot_sha256":"5e4c80c66af0ca401df3410dd2b1dc3fd6c8d168c508588a3c422e716c45b319","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Finetuning can significantly modify the behavior of large language models, including introducing harmful or unsafe behaviors. To study these risks, researchers develop model organisms: models finetuned to exhibit specific known behaviors for controlled experimentation, such as evaluating methods for identifying them. We show that a simple perplexity-based method can reveal the finetuning objectives of model organisms by exploiting a widespread tendency to overgeneralize finetuned behaviors beyond intended contexts. We generate diverse completions from the finetuned model using short random pre","authors_text":"Dan Wilhelm, Luca Baroni, Mohammed Abu Baker","cross_cats":["cs.AI"],"headline":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-01T18:00:55Z","title":"Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.00994","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-09T18:43:49.422700Z","id":"1f349c81-ec89-41f7-be14-e19108758f33","model_set":{"reader":"grok-4.3"},"one_line_summary":"Perplexity gaps between finetuned and reference models on random-prefill completions often reveal the original finetuning objectives across diverse model organisms.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms.","strongest_claim":"a simple perplexity-based method can surface finetuning objectives from model organisms by leveraging their tendency to overgeneralize their finetuned behaviors beyond the intended context","weakest_assumption":"that finetuned models reliably overgeneralize their training objectives to unrelated contexts generated from random prefills, producing detectable perplexity gaps against a reference model"}},"verdict_id":"1f349c81-ec89-41f7-be14-e19108758f33"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:36b766eb039e6d2a9e621b9fb7cb6204ce5a6984bb5c9c73e409f066db20c1dc","target":"record","created_at":"2026-06-30T02:18:08Z","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":"98133f29ab394c6fc1eb1e7453dce3e825395a16bc5f801aae67f197ba48a623","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-01T18:00:55Z","title_canon_sha256":"bbeb7847e2c0218b82d28e816a5d6a57d3dc09a8f6399c5e4101c304f0ba6784"},"schema_version":"1.0","source":{"id":"2605.00994","kind":"arxiv","version":2}},"canonical_sha256":"5ef7169a980da5dff973df87c34edbb2f1730eddd9597ab8b78e8106ed54ac93","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"5ef7169a980da5dff973df87c34edbb2f1730eddd9597ab8b78e8106ed54ac93","first_computed_at":"2026-06-30T02:18:08.784645Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-30T02:18:08.784645Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"FDnnDkKlHj6PqQ1Px4rMGbyqRz8Obv2WAaArrr51lg6kNxhYctWajb14MeXahWVmK0Zksn8rMnw1HfP628mpDQ==","signature_status":"signed_v1","signed_at":"2026-06-30T02:18:08.785140Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.00994","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:36b766eb039e6d2a9e621b9fb7cb6204ce5a6984bb5c9c73e409f066db20c1dc","sha256:6f1ed3a768707114d359baae71bf06eec6016af20ad31b8aa918afe1b1890e05"],"state_sha256":"9eefa0e06aab29b2c9501209e7e802d750f38cec4814a2903260274f5940589a"}