{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3G2V26Z7LXDNC2EQZMIXIA4CPH","short_pith_number":"pith:3G2V26Z7","schema_version":"1.0","canonical_sha256":"d9b55d7b3f5dc6d16890cb1174038279ec00ecabcfaef4d284ee2fe2ebf54636","source":{"kind":"arxiv","id":"2312.00027","version":2},"attestation_state":"computed","paper":{"title":"Stealthy and Persistent Unalignment on Large Language Models via Backdoor Injections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Bochuan Cao, Jinghui Chen, Yuanpu Cao","submitted_at":"2023-11-15T23:52:05Z","abstract_excerpt":"Recent developments in Large Language Models (LLMs) have manifested significant advancements. To facilitate safeguards against malicious exploitation, a body of research has concentrated on aligning LLMs with human preferences and inhibiting their generation of inappropriate content. Unfortunately, such alignments are often vulnerable: fine-tuning with a minimal amount of harmful data can easily unalign the target LLM. While being effective, such fine-tuning-based unalignment approaches also have their own limitations: (1) non-stealthiness, after fine-tuning, safety audits or red-teaming can e"},"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":"2312.00027","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2023-11-15T23:52:05Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"45409989427b3a8cebf04222837743cea572f06f9ead0485167a97001b48e8e7","abstract_canon_sha256":"8ba83cfde49642284e219ebdd84bb1744397cb38c708e02ccb1d247f65dc4372"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:29:18.866497Z","signature_b64":"+EfrQUt4DaME1sscIw0d0B7eTMecFzTLoBU3rquXlkINi53GIw0XJJyZ3sDwHS9+PnFZZHDapbKjy9QYUGaKBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9b55d7b3f5dc6d16890cb1174038279ec00ecabcfaef4d284ee2fe2ebf54636","last_reissued_at":"2026-07-05T08:29:18.866004Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:29:18.866004Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stealthy and Persistent Unalignment on Large Language Models via Backdoor Injections","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Bochuan Cao, Jinghui Chen, Yuanpu Cao","submitted_at":"2023-11-15T23:52:05Z","abstract_excerpt":"Recent developments in Large Language Models (LLMs) have manifested significant advancements. To facilitate safeguards against malicious exploitation, a body of research has concentrated on aligning LLMs with human preferences and inhibiting their generation of inappropriate content. Unfortunately, such alignments are often vulnerable: fine-tuning with a minimal amount of harmful data can easily unalign the target LLM. While being effective, such fine-tuning-based unalignment approaches also have their own limitations: (1) non-stealthiness, after fine-tuning, safety audits or red-teaming can e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.00027","kind":"arxiv","version":2},"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/2312.00027/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":"2312.00027","created_at":"2026-07-05T08:29:18.866060+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.00027v2","created_at":"2026-07-05T08:29:18.866060+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.00027","created_at":"2026-07-05T08:29:18.866060+00:00"},{"alias_kind":"pith_short_12","alias_value":"3G2V26Z7LXDN","created_at":"2026-07-05T08:29:18.866060+00:00"},{"alias_kind":"pith_short_16","alias_value":"3G2V26Z7LXDNC2EQ","created_at":"2026-07-05T08:29:18.866060+00:00"},{"alias_kind":"pith_short_8","alias_value":"3G2V26Z7","created_at":"2026-07-05T08:29:18.866060+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.07287","citing_title":"Paladin: Defending LLM-enabled Phishing Emails with a New Trigger-Tag Paradigm","ref_index":161,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH","json":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH.json","graph_json":"https://pith.science/api/pith-number/3G2V26Z7LXDNC2EQZMIXIA4CPH/graph.json","events_json":"https://pith.science/api/pith-number/3G2V26Z7LXDNC2EQZMIXIA4CPH/events.json","paper":"https://pith.science/paper/3G2V26Z7"},"agent_actions":{"view_html":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH","download_json":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH.json","view_paper":"https://pith.science/paper/3G2V26Z7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.00027&json=true","fetch_graph":"https://pith.science/api/pith-number/3G2V26Z7LXDNC2EQZMIXIA4CPH/graph.json","fetch_events":"https://pith.science/api/pith-number/3G2V26Z7LXDNC2EQZMIXIA4CPH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH/action/storage_attestation","attest_author":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH/action/author_attestation","sign_citation":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH/action/citation_signature","submit_replication":"https://pith.science/pith/3G2V26Z7LXDNC2EQZMIXIA4CPH/action/replication_record"}},"created_at":"2026-07-05T08:29:18.866060+00:00","updated_at":"2026-07-05T08:29:18.866060+00:00"}