{"paper":{"title":"Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dan Wilhelm, Luca Baroni, Mohammed Abu Baker","submitted_at":"2026-05-01T18:00:55Z","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"},"claims":{"count":4,"items":[{"kind":"strongest_claim","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","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"that finetuned models reliably overgeneralize their training objectives to unrelated contexts generated from random prefills, producing detectable perplexity gaps against a reference model","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Perplexity gaps between finetuned and reference models on random-prefill completions often reveal the original finetuning objectives across diverse model organisms.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"92ed26ec09b935a7177b025b6539c412aabe204323ca7a3a900a9dbf28128d73"},"source":{"id":"2605.00994","kind":"arxiv","version":2},"verdict":{"id":"1f349c81-ec89-41f7-be14-e19108758f33","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T18:43:49.422700Z","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","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","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","pith_extraction_headline":"Perplexity differencing on completions from random prefills surfaces finetuning objectives in most model organisms."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00994/integrity.json","findings":[],"available":true,"detectors_run":[{"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","findings_count":0}],"snapshot_sha256":"5e4c80c66af0ca401df3410dd2b1dc3fd6c8d168c508588a3c422e716c45b319"},"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"}