{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:O62NKL3TWSREUP2MYTITDVAHAB","short_pith_number":"pith:O62NKL3T","schema_version":"1.0","canonical_sha256":"77b4d52f73b4a24a3f4cc4d131d4070042a49c0dea07da89a436b9ba06491996","source":{"kind":"arxiv","id":"2410.08827","version":3},"attestation_state":"computed","paper":{"title":"Do Unlearning Methods Remove Information from Language Model Weights?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aghyad Deeb, Fabien Roger","submitted_at":"2024-10-11T14:06:58Z","abstract_excerpt":"Large Language Models' knowledge of how to perform cyber-security attacks, create bioweapons, and manipulate humans poses risks of misuse. Previous work has proposed methods to unlearn this knowledge. Historically, it has been unclear whether unlearning techniques are removing information from the model weights or just making it harder to access. To disentangle these two objectives, we propose an adversarial evaluation method to test for the removal of information from model weights: we give an attacker access to some facts that were supposed to be removed, and using those, the attacker tries "},"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":"2410.08827","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-11T14:06:58Z","cross_cats_sorted":[],"title_canon_sha256":"d11e9dd72928f158557d5c39adb07c66eda7ded1d7ff5015f9e7fbc5bd3ff2ff","abstract_canon_sha256":"3514d1c8897ecbe088e9888b300b02eb4761296392f3dfae71193631ca4386b7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:10:44.863734Z","signature_b64":"8kRTnf30ebpe/lSGFGwbJXfuL2XwsZ2du+03R9RwMTMk5wnUxQNoTmhSVjLjBuo/AnAfEnt0/zW/66VatCLRCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"77b4d52f73b4a24a3f4cc4d131d4070042a49c0dea07da89a436b9ba06491996","last_reissued_at":"2026-07-05T10:10:44.863309Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:10:44.863309Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Do Unlearning Methods Remove Information from Language Model Weights?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aghyad Deeb, Fabien Roger","submitted_at":"2024-10-11T14:06:58Z","abstract_excerpt":"Large Language Models' knowledge of how to perform cyber-security attacks, create bioweapons, and manipulate humans poses risks of misuse. Previous work has proposed methods to unlearn this knowledge. Historically, it has been unclear whether unlearning techniques are removing information from the model weights or just making it harder to access. To disentangle these two objectives, we propose an adversarial evaluation method to test for the removal of information from model weights: we give an attacker access to some facts that were supposed to be removed, and using those, the attacker tries "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.08827","kind":"arxiv","version":3},"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/2410.08827/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":"2410.08827","created_at":"2026-07-05T10:10:44.863381+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.08827v3","created_at":"2026-07-05T10:10:44.863381+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.08827","created_at":"2026-07-05T10:10:44.863381+00:00"},{"alias_kind":"pith_short_12","alias_value":"O62NKL3TWSRE","created_at":"2026-07-05T10:10:44.863381+00:00"},{"alias_kind":"pith_short_16","alias_value":"O62NKL3TWSREUP2M","created_at":"2026-07-05T10:10:44.863381+00:00"},{"alias_kind":"pith_short_8","alias_value":"O62NKL3T","created_at":"2026-07-05T10:10:44.863381+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17168","citing_title":"RepSelect: Robust LLM Unlearning via Representation Selectivity","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09559","citing_title":"Safe-RULE: Safe Reinforcement UnLEarning","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03695","citing_title":"Don't Forget Your Embeddings: Robust Knowledge Erasure via Precise Editing of Embeddings","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26526","citing_title":"Open-Weight LLM Fine-Tuning Defenses are Susceptible to Simple Attacks","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2501.19202","citing_title":"Improving LLM Unlearning Robustness via Random Perturbations","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2510.00761","citing_title":"Downgrade to Upgrade: Optimizer Simplification Enhances Robustness in LLM Unlearning","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11685","citing_title":"Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09391","citing_title":"Efficient Unlearning through Maximizing Relearning Convergence Delay","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07962","citing_title":"Is your algorithm unlearning or untraining?","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13438","citing_title":"WIN-U: Woodbury-Informed Newton-Unlearning as a retain-free Machine Unlearning Framework","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB","json":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB.json","graph_json":"https://pith.science/api/pith-number/O62NKL3TWSREUP2MYTITDVAHAB/graph.json","events_json":"https://pith.science/api/pith-number/O62NKL3TWSREUP2MYTITDVAHAB/events.json","paper":"https://pith.science/paper/O62NKL3T"},"agent_actions":{"view_html":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB","download_json":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB.json","view_paper":"https://pith.science/paper/O62NKL3T","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.08827&json=true","fetch_graph":"https://pith.science/api/pith-number/O62NKL3TWSREUP2MYTITDVAHAB/graph.json","fetch_events":"https://pith.science/api/pith-number/O62NKL3TWSREUP2MYTITDVAHAB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB/action/storage_attestation","attest_author":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB/action/author_attestation","sign_citation":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB/action/citation_signature","submit_replication":"https://pith.science/pith/O62NKL3TWSREUP2MYTITDVAHAB/action/replication_record"}},"created_at":"2026-07-05T10:10:44.863381+00:00","updated_at":"2026-07-05T10:10:44.863381+00:00"}