{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YF5WLUAI3XV267QYVMQPC66D44","short_pith_number":"pith:YF5WLUAI","schema_version":"1.0","canonical_sha256":"c17b65d008ddebaf7e18ab20f17bc3e724a3bc214f316c7ee69fcd573087cb65","source":{"kind":"arxiv","id":"2502.05945","version":3},"attestation_state":"computed","paper":{"title":"Head-Specific Intervention Can Induce Misaligned AI Coordination in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Annalisa Riccardi, Paul Darm","submitted_at":"2025-02-09T16:11:57Z","abstract_excerpt":"Robust alignment guardrails for large language models (LLMs) are becoming increasingly important with their widespread application. In contrast to previous studies, we demonstrate that inference-time activation interventions can bypass safety alignments and effectively steer model generations towards harmful AI coordination. Our method applies fine-grained interventions at specific attention heads, which we identify by probing each head in a simple binary choice task. We then show that interventions on these heads generalise to the open-ended generation setting, effectively circumventing safet"},"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":"2502.05945","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-09T16:11:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"ec9c5ec722a368a18ab44e82ad5877cdffcedf32ba243a2659d6145e6629a3e3","abstract_canon_sha256":"bb0d41cf06306d75a67a508199cb735eac6043132d534df8cdff6d6c16a276c0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:58:24.700616Z","signature_b64":"HzEoB/ku1GfJfo9sIWDS0UYDcDbN3HnhVzXV5YOVkO78DPXAnWPVCknx0wzxJ4nIhzdJOvYSlWE6MVAbvx5WBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c17b65d008ddebaf7e18ab20f17bc3e724a3bc214f316c7ee69fcd573087cb65","last_reissued_at":"2026-07-05T11:58:24.700177Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:58:24.700177Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Head-Specific Intervention Can Induce Misaligned AI Coordination in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Annalisa Riccardi, Paul Darm","submitted_at":"2025-02-09T16:11:57Z","abstract_excerpt":"Robust alignment guardrails for large language models (LLMs) are becoming increasingly important with their widespread application. In contrast to previous studies, we demonstrate that inference-time activation interventions can bypass safety alignments and effectively steer model generations towards harmful AI coordination. Our method applies fine-grained interventions at specific attention heads, which we identify by probing each head in a simple binary choice task. We then show that interventions on these heads generalise to the open-ended generation setting, effectively circumventing safet"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.05945","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/2502.05945/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":"2502.05945","created_at":"2026-07-05T11:58:24.700233+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.05945v3","created_at":"2026-07-05T11:58:24.700233+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.05945","created_at":"2026-07-05T11:58:24.700233+00:00"},{"alias_kind":"pith_short_12","alias_value":"YF5WLUAI3XV2","created_at":"2026-07-05T11:58:24.700233+00:00"},{"alias_kind":"pith_short_16","alias_value":"YF5WLUAI3XV267QY","created_at":"2026-07-05T11:58:24.700233+00:00"},{"alias_kind":"pith_short_8","alias_value":"YF5WLUAI","created_at":"2026-07-05T11:58:24.700233+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/YF5WLUAI3XV267QYVMQPC66D44","json":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44.json","graph_json":"https://pith.science/api/pith-number/YF5WLUAI3XV267QYVMQPC66D44/graph.json","events_json":"https://pith.science/api/pith-number/YF5WLUAI3XV267QYVMQPC66D44/events.json","paper":"https://pith.science/paper/YF5WLUAI"},"agent_actions":{"view_html":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44","download_json":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44.json","view_paper":"https://pith.science/paper/YF5WLUAI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.05945&json=true","fetch_graph":"https://pith.science/api/pith-number/YF5WLUAI3XV267QYVMQPC66D44/graph.json","fetch_events":"https://pith.science/api/pith-number/YF5WLUAI3XV267QYVMQPC66D44/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44/action/storage_attestation","attest_author":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44/action/author_attestation","sign_citation":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44/action/citation_signature","submit_replication":"https://pith.science/pith/YF5WLUAI3XV267QYVMQPC66D44/action/replication_record"}},"created_at":"2026-07-05T11:58:24.700233+00:00","updated_at":"2026-07-05T11:58:24.700233+00:00"}