{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TUGC3AGIVIIPE3JENBUF7ZDTAQ","short_pith_number":"pith:TUGC3AGI","schema_version":"1.0","canonical_sha256":"9d0c2d80c8aa10f26d2468685fe473040c8676eed5f9fd5b17da97a531d455b5","source":{"kind":"arxiv","id":"2505.16559","version":1},"attestation_state":"computed","paper":{"title":"CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Baolei Zhang, Biao Yi, Lihai Nie, Li Shen, Tiansheng Huang, Tong Li, Zheli Liu","submitted_at":"2025-05-22T11:47:08Z","abstract_excerpt":"Fine-tuning-as-a-service, while commercially successful for Large Language Model (LLM) providers, exposes models to harmful fine-tuning attacks. As a widely explored defense paradigm against such attacks, unlearning attempts to remove malicious knowledge from LLMs, thereby essentially preventing them from being used to perform malicious tasks. However, we highlight a critical flaw: the powerful general adaptability of LLMs allows them to easily bypass selective unlearning by rapidly relearning or repurposing their capabilities for harmful tasks. To address this fundamental limitation, we propo"},"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":"2505.16559","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-05-22T11:47:08Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"5ea180060627bc4f9d45d4e3d8896deb0ccbe96b3444a3e3979b5d34801a968b","abstract_canon_sha256":"420cdc28c434f7bad5cdcffc581d586268528fe548c2589ad5325515e92f2a3c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:07:37.234983Z","signature_b64":"KC8zmIiqy65o+qoKT3KhcIUQxjP3UzoM7UfI/woJj2HDoayAU4KYtcS5vZursWvlMMWn38mSsNd/n43JPVZ2CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9d0c2d80c8aa10f26d2468685fe473040c8676eed5f9fd5b17da97a531d455b5","last_reissued_at":"2026-07-05T11:07:37.234465Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:07:37.234465Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Baolei Zhang, Biao Yi, Lihai Nie, Li Shen, Tiansheng Huang, Tong Li, Zheli Liu","submitted_at":"2025-05-22T11:47:08Z","abstract_excerpt":"Fine-tuning-as-a-service, while commercially successful for Large Language Model (LLM) providers, exposes models to harmful fine-tuning attacks. As a widely explored defense paradigm against such attacks, unlearning attempts to remove malicious knowledge from LLMs, thereby essentially preventing them from being used to perform malicious tasks. However, we highlight a critical flaw: the powerful general adaptability of LLMs allows them to easily bypass selective unlearning by rapidly relearning or repurposing their capabilities for harmful tasks. To address this fundamental limitation, we propo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.16559","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/2505.16559/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":"2505.16559","created_at":"2026-07-05T11:07:37.234535+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.16559v1","created_at":"2026-07-05T11:07:37.234535+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.16559","created_at":"2026-07-05T11:07:37.234535+00:00"},{"alias_kind":"pith_short_12","alias_value":"TUGC3AGIVIIP","created_at":"2026-07-05T11:07:37.234535+00:00"},{"alias_kind":"pith_short_16","alias_value":"TUGC3AGIVIIPE3JE","created_at":"2026-07-05T11:07:37.234535+00:00"},{"alias_kind":"pith_short_8","alias_value":"TUGC3AGI","created_at":"2026-07-05T11:07:37.234535+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.14605","citing_title":"One Step to the Side: Why Defenses Against Malicious Finetuning Fail Under Adaptive Adversaries","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2508.20697","citing_title":"Token Buncher: Shielding LLMs from Harmful Reinforcement Learning Fine-Tuning","ref_index":61,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ","json":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ.json","graph_json":"https://pith.science/api/pith-number/TUGC3AGIVIIPE3JENBUF7ZDTAQ/graph.json","events_json":"https://pith.science/api/pith-number/TUGC3AGIVIIPE3JENBUF7ZDTAQ/events.json","paper":"https://pith.science/paper/TUGC3AGI"},"agent_actions":{"view_html":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ","download_json":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ.json","view_paper":"https://pith.science/paper/TUGC3AGI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.16559&json=true","fetch_graph":"https://pith.science/api/pith-number/TUGC3AGIVIIPE3JENBUF7ZDTAQ/graph.json","fetch_events":"https://pith.science/api/pith-number/TUGC3AGIVIIPE3JENBUF7ZDTAQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ/action/storage_attestation","attest_author":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ/action/author_attestation","sign_citation":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ/action/citation_signature","submit_replication":"https://pith.science/pith/TUGC3AGIVIIPE3JENBUF7ZDTAQ/action/replication_record"}},"created_at":"2026-07-05T11:07:37.234535+00:00","updated_at":"2026-07-05T11:07:37.234535+00:00"}