{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:KP467U67NEJXI4SXSCSPQ5KUEE","short_pith_number":"pith:KP467U67","schema_version":"1.0","canonical_sha256":"53f9efd3df691374725790a4f875542131f38a4adf12a0a05fabd68c966e9558","source":{"kind":"arxiv","id":"2311.00117","version":3},"attestation_state":"computed","paper":{"title":"BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Charlie Rogers-Smith, Jeffrey Ladish, Pranav Gade, Simon Lermen","submitted_at":"2023-10-31T19:45:15Z","abstract_excerpt":"Llama 2-Chat is a collection of large language models that Meta developed and released to the public. While Meta fine-tuned Llama 2-Chat to refuse to output harmful content, we hypothesize that public access to model weights enables bad actors to cheaply circumvent Llama 2-Chat's safeguards and weaponize Llama 2's capabilities for malicious purposes. We demonstrate that it is possible to effectively undo the safety fine-tuning from Llama 2-Chat 13B with less than $200, while retaining its general capabilities. Our results demonstrate that safety-fine tuning is ineffective at preventing misuse "},"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":"2311.00117","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2023-10-31T19:45:15Z","cross_cats_sorted":[],"title_canon_sha256":"340ba3ceb079cdf76d68b6be9897bcafa037ad856579a1a7763a1cf2a12539a5","abstract_canon_sha256":"f46879deef3cbf79e062e18921a1468388adf2f7d4d807314bc2ba582fe432a7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:23:57.262870Z","signature_b64":"buHa4B9FshVTvFS7SXj+5SSCPHMeYEL0asWGvCuixzcVAlARflf0Bq0ZEsU15dziWpdK0lsB38Pv0NHkDhiKDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"53f9efd3df691374725790a4f875542131f38a4adf12a0a05fabd68c966e9558","last_reissued_at":"2026-07-05T08:23:57.262447Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:23:57.262447Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BadLlama: cheaply removing safety fine-tuning from Llama 2-Chat 13B","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Charlie Rogers-Smith, Jeffrey Ladish, Pranav Gade, Simon Lermen","submitted_at":"2023-10-31T19:45:15Z","abstract_excerpt":"Llama 2-Chat is a collection of large language models that Meta developed and released to the public. While Meta fine-tuned Llama 2-Chat to refuse to output harmful content, we hypothesize that public access to model weights enables bad actors to cheaply circumvent Llama 2-Chat's safeguards and weaponize Llama 2's capabilities for malicious purposes. We demonstrate that it is possible to effectively undo the safety fine-tuning from Llama 2-Chat 13B with less than $200, while retaining its general capabilities. Our results demonstrate that safety-fine tuning is ineffective at preventing misuse "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.00117","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/2311.00117/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":"2311.00117","created_at":"2026-07-05T08:23:57.262514+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.00117v3","created_at":"2026-07-05T08:23:57.262514+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.00117","created_at":"2026-07-05T08:23:57.262514+00:00"},{"alias_kind":"pith_short_12","alias_value":"KP467U67NEJX","created_at":"2026-07-05T08:23:57.262514+00:00"},{"alias_kind":"pith_short_16","alias_value":"KP467U67NEJXI4SX","created_at":"2026-07-05T08:23:57.262514+00:00"},{"alias_kind":"pith_short_8","alias_value":"KP467U67","created_at":"2026-07-05T08:23:57.262514+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.00160","citing_title":"DataShield: Safety-degrading Data Filtering for LLM Benign Instruction Fine-Tuning","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16471","citing_title":"From AI-Generated Content to Agentic Action: Security and Safety Threats in Generative AI","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2506.06414","citing_title":"Benchmarking Misuse Mitigation Against Covert Adversaries","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE","json":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE.json","graph_json":"https://pith.science/api/pith-number/KP467U67NEJXI4SXSCSPQ5KUEE/graph.json","events_json":"https://pith.science/api/pith-number/KP467U67NEJXI4SXSCSPQ5KUEE/events.json","paper":"https://pith.science/paper/KP467U67"},"agent_actions":{"view_html":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE","download_json":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE.json","view_paper":"https://pith.science/paper/KP467U67","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.00117&json=true","fetch_graph":"https://pith.science/api/pith-number/KP467U67NEJXI4SXSCSPQ5KUEE/graph.json","fetch_events":"https://pith.science/api/pith-number/KP467U67NEJXI4SXSCSPQ5KUEE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE/action/storage_attestation","attest_author":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE/action/author_attestation","sign_citation":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE/action/citation_signature","submit_replication":"https://pith.science/pith/KP467U67NEJXI4SXSCSPQ5KUEE/action/replication_record"}},"created_at":"2026-07-05T08:23:57.262514+00:00","updated_at":"2026-07-05T08:23:57.262514+00:00"}