{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HVL6X3YFBVHQ7BT7MNVUPMLIPB","short_pith_number":"pith:HVL6X3YF","schema_version":"1.0","canonical_sha256":"3d57ebef050d4f0f867f636b47b168784bcb7b4a78ad491d1956edee13cd6672","source":{"kind":"arxiv","id":"2602.12418","version":2},"attestation_state":"computed","paper":{"title":"Sparse Autoencoders are Capable LLM Jailbreak Mitigators","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Arno Blaas, Jacopo Cortellazzi, Javier Abad, Pau Rodriguez, Xavier Suau, Yannick Assogba","submitted_at":"2026-02-12T21:17:32Z","abstract_excerpt":"Jailbreak attacks remain a persistent threat to large language model safety. We propose Context-Conditioned Delta Steering (CC-Delta), an SAE-based defense that identifies jailbreak-relevant sparse features by comparing token-level representations of the same harmful request with and without jailbreak context. Using paired harmful/jailbreak prompts, CC-Delta selects features via statistical testing and applies inference-time mean-shift steering in SAE latent space. Across four aligned instruction-tuned models and twelve jailbreak attacks, CC-Delta achieves comparable or better safety-utility t"},"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":"2602.12418","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-02-12T21:17:32Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"f9da5fe5a913d9fa36237ece97b4b63d4d513db81b2653805cba4fab69d2c472","abstract_canon_sha256":"3fde68b965a33163d13b16e170ff32a27bfaac2613f8f6699d95e4902fa9e6fc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:16:28.305394Z","signature_b64":"NIWuafrWNEu1/8WqDLPs52yYRvohrsUqzMeP40u6gQHZd0OhrKsH/Y+rn8eft7eTRWBLabh6Kb/D+V5ixr0mAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3d57ebef050d4f0f867f636b47b168784bcb7b4a78ad491d1956edee13cd6672","last_reissued_at":"2026-06-30T01:16:28.304857Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:16:28.304857Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sparse Autoencoders are Capable LLM Jailbreak Mitigators","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Arno Blaas, Jacopo Cortellazzi, Javier Abad, Pau Rodriguez, Xavier Suau, Yannick Assogba","submitted_at":"2026-02-12T21:17:32Z","abstract_excerpt":"Jailbreak attacks remain a persistent threat to large language model safety. We propose Context-Conditioned Delta Steering (CC-Delta), an SAE-based defense that identifies jailbreak-relevant sparse features by comparing token-level representations of the same harmful request with and without jailbreak context. Using paired harmful/jailbreak prompts, CC-Delta selects features via statistical testing and applies inference-time mean-shift steering in SAE latent space. Across four aligned instruction-tuned models and twelve jailbreak attacks, CC-Delta achieves comparable or better safety-utility t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.12418","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/2602.12418/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":"2602.12418","created_at":"2026-06-30T01:16:28.304916+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.12418v2","created_at":"2026-06-30T01:16:28.304916+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.12418","created_at":"2026-06-30T01:16:28.304916+00:00"},{"alias_kind":"pith_short_12","alias_value":"HVL6X3YFBVHQ","created_at":"2026-06-30T01:16:28.304916+00:00"},{"alias_kind":"pith_short_16","alias_value":"HVL6X3YFBVHQ7BT7","created_at":"2026-06-30T01:16:28.304916+00:00"},{"alias_kind":"pith_short_8","alias_value":"HVL6X3YF","created_at":"2026-06-30T01:16:28.304916+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/HVL6X3YFBVHQ7BT7MNVUPMLIPB","json":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB.json","graph_json":"https://pith.science/api/pith-number/HVL6X3YFBVHQ7BT7MNVUPMLIPB/graph.json","events_json":"https://pith.science/api/pith-number/HVL6X3YFBVHQ7BT7MNVUPMLIPB/events.json","paper":"https://pith.science/paper/HVL6X3YF"},"agent_actions":{"view_html":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB","download_json":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB.json","view_paper":"https://pith.science/paper/HVL6X3YF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.12418&json=true","fetch_graph":"https://pith.science/api/pith-number/HVL6X3YFBVHQ7BT7MNVUPMLIPB/graph.json","fetch_events":"https://pith.science/api/pith-number/HVL6X3YFBVHQ7BT7MNVUPMLIPB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB/action/storage_attestation","attest_author":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB/action/author_attestation","sign_citation":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB/action/citation_signature","submit_replication":"https://pith.science/pith/HVL6X3YFBVHQ7BT7MNVUPMLIPB/action/replication_record"}},"created_at":"2026-06-30T01:16:28.304916+00:00","updated_at":"2026-06-30T01:16:28.304916+00:00"}