{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OP3IYXKFDOKTFLG4J2ZPPPYGQX","short_pith_number":"pith:OP3IYXKF","schema_version":"1.0","canonical_sha256":"73f68c5d451b9532acdc4eb2f7bf0685e43bdfafe5e4a9d177c440d35221a2af","source":{"kind":"arxiv","id":"2407.01376","version":1},"attestation_state":"computed","paper":{"title":"Badllama 3: removing safety finetuning from Llama 3 in minutes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Dmitrii Volkov","submitted_at":"2024-07-01T15:29:45Z","abstract_excerpt":"We show that extensive LLM safety fine-tuning is easily subverted when an attacker has access to model weights. We evaluate three state-of-the-art fine-tuning methods-QLoRA, ReFT, and Ortho-and show how algorithmic advances enable constant jailbreaking performance with cuts in FLOPs and optimisation power. We strip safety fine-tuning from Llama 3 8B in one minute and Llama 3 70B in 30 minutes on a single GPU, and sketch ways to reduce this further."},"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":"2407.01376","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-07-01T15:29:45Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"title_canon_sha256":"2c267dc89c6354ee41a49e051ca439c2992ddb9f74da3ac84b14d3e8247019a0","abstract_canon_sha256":"cedad507bc1c0974ec0a4447aa00607f8247cc2c3027526c8823d1da390f21b1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:38:47.120348Z","signature_b64":"sAwKgZwnktbe45K9AATT1K4dXmY5GtwQDODwElXN3WGMMhIlNFeOP2yyRzUqzc2SwoBcrFgILuGmUFAJnsVdBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"73f68c5d451b9532acdc4eb2f7bf0685e43bdfafe5e4a9d177c440d35221a2af","last_reissued_at":"2026-07-05T08:38:47.119785Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:38:47.119785Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Badllama 3: removing safety finetuning from Llama 3 in minutes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Dmitrii Volkov","submitted_at":"2024-07-01T15:29:45Z","abstract_excerpt":"We show that extensive LLM safety fine-tuning is easily subverted when an attacker has access to model weights. We evaluate three state-of-the-art fine-tuning methods-QLoRA, ReFT, and Ortho-and show how algorithmic advances enable constant jailbreaking performance with cuts in FLOPs and optimisation power. We strip safety fine-tuning from Llama 3 8B in one minute and Llama 3 70B in 30 minutes on a single GPU, and sketch ways to reduce this further."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.01376","kind":"arxiv","version":1},"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/2407.01376/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":"2407.01376","created_at":"2026-07-05T08:38:47.119838+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.01376v1","created_at":"2026-07-05T08:38:47.119838+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.01376","created_at":"2026-07-05T08:38:47.119838+00:00"},{"alias_kind":"pith_short_12","alias_value":"OP3IYXKFDOKT","created_at":"2026-07-05T08:38:47.119838+00:00"},{"alias_kind":"pith_short_16","alias_value":"OP3IYXKFDOKTFLG4","created_at":"2026-07-05T08:38:47.119838+00:00"},{"alias_kind":"pith_short_8","alias_value":"OP3IYXKF","created_at":"2026-07-05T08:38:47.119838+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.17109","citing_title":"Mitigating Cyber Risk in the Age of Open-Weight LLMs: Policy Gaps and Technical Realities","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX","json":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX.json","graph_json":"https://pith.science/api/pith-number/OP3IYXKFDOKTFLG4J2ZPPPYGQX/graph.json","events_json":"https://pith.science/api/pith-number/OP3IYXKFDOKTFLG4J2ZPPPYGQX/events.json","paper":"https://pith.science/paper/OP3IYXKF"},"agent_actions":{"view_html":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX","download_json":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX.json","view_paper":"https://pith.science/paper/OP3IYXKF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.01376&json=true","fetch_graph":"https://pith.science/api/pith-number/OP3IYXKFDOKTFLG4J2ZPPPYGQX/graph.json","fetch_events":"https://pith.science/api/pith-number/OP3IYXKFDOKTFLG4J2ZPPPYGQX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX/action/storage_attestation","attest_author":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX/action/author_attestation","sign_citation":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX/action/citation_signature","submit_replication":"https://pith.science/pith/OP3IYXKFDOKTFLG4J2ZPPPYGQX/action/replication_record"}},"created_at":"2026-07-05T08:38:47.119838+00:00","updated_at":"2026-07-05T08:38:47.119838+00:00"}