{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BR6IIYKLCTLW36TO2QEFOBVHTI","short_pith_number":"pith:BR6IIYKL","schema_version":"1.0","canonical_sha256":"0c7c84614b14d76dfa6ed4085706a79a3347fcc8bc68856ba321eb4bc89178bf","source":{"kind":"arxiv","id":"2506.10020","version":1},"attestation_state":"computed","paper":{"title":"From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Hyunbin Jin, Kyubyung Chae, Taesup Kim","submitted_at":"2025-06-07T08:19:01Z","abstract_excerpt":"Safely aligning large language models (LLMs) often demands extensive human-labeled preference data, a process that's both costly and time-consuming. While synthetic data offers a promising alternative, current methods frequently rely on complex iterative prompting or auxiliary models. To address this, we introduce Refusal-Aware Adaptive Injection (RAAI), a straightforward, training-free, and model-agnostic framework that repurposes LLM attack techniques. RAAI works by detecting internal refusal signals and adaptively injecting predefined phrases to elicit harmful, yet fluent, completions. Our "},"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":"2506.10020","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-06-07T08:19:01Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"db86fe356bd8d592fab109186f9d7867f399a2a65a6afc2f013b3c1ae7f8397a","abstract_canon_sha256":"87ea1f924391097357bd775ecd7194c91290b736455b16857c8da94f69f74736"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:38.608827Z","signature_b64":"Rf1MTNPlo+USR1d3beadkqDhZMSfb91PZcxZeoW2zWVWIcdvzQ1zxS5sGUJPiiBLd96cCLXEbNmFiIstoN/TCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0c7c84614b14d76dfa6ed4085706a79a3347fcc8bc68856ba321eb4bc89178bf","last_reissued_at":"2026-07-05T11:20:38.608375Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:38.608375Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CR","authors_text":"Hyunbin Jin, Kyubyung Chae, Taesup Kim","submitted_at":"2025-06-07T08:19:01Z","abstract_excerpt":"Safely aligning large language models (LLMs) often demands extensive human-labeled preference data, a process that's both costly and time-consuming. While synthetic data offers a promising alternative, current methods frequently rely on complex iterative prompting or auxiliary models. To address this, we introduce Refusal-Aware Adaptive Injection (RAAI), a straightforward, training-free, and model-agnostic framework that repurposes LLM attack techniques. RAAI works by detecting internal refusal signals and adaptively injecting predefined phrases to elicit harmful, yet fluent, completions. Our "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.10020","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/2506.10020/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":"2506.10020","created_at":"2026-07-05T11:20:38.608441+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.10020v1","created_at":"2026-07-05T11:20:38.608441+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.10020","created_at":"2026-07-05T11:20:38.608441+00:00"},{"alias_kind":"pith_short_12","alias_value":"BR6IIYKLCTLW","created_at":"2026-07-05T11:20:38.608441+00:00"},{"alias_kind":"pith_short_16","alias_value":"BR6IIYKLCTLW36TO","created_at":"2026-07-05T11:20:38.608441+00:00"},{"alias_kind":"pith_short_8","alias_value":"BR6IIYKL","created_at":"2026-07-05T11:20:38.608441+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI","json":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI.json","graph_json":"https://pith.science/api/pith-number/BR6IIYKLCTLW36TO2QEFOBVHTI/graph.json","events_json":"https://pith.science/api/pith-number/BR6IIYKLCTLW36TO2QEFOBVHTI/events.json","paper":"https://pith.science/paper/BR6IIYKL"},"agent_actions":{"view_html":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI","download_json":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI.json","view_paper":"https://pith.science/paper/BR6IIYKL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.10020&json=true","fetch_graph":"https://pith.science/api/pith-number/BR6IIYKLCTLW36TO2QEFOBVHTI/graph.json","fetch_events":"https://pith.science/api/pith-number/BR6IIYKLCTLW36TO2QEFOBVHTI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI/action/storage_attestation","attest_author":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI/action/author_attestation","sign_citation":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI/action/citation_signature","submit_replication":"https://pith.science/pith/BR6IIYKLCTLW36TO2QEFOBVHTI/action/replication_record"}},"created_at":"2026-07-05T11:20:38.608441+00:00","updated_at":"2026-07-05T11:20:38.608441+00:00"}