{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XSJ3XHZK2ZPSUZPRDOHE4R7AGB","short_pith_number":"pith:XSJ3XHZK","schema_version":"1.0","canonical_sha256":"bc93bb9f2ad65f2a65f11b8e4e47e03042df5ceae1509353b26edf3f142a2f58","source":{"kind":"arxiv","id":"2605.24535","version":1},"attestation_state":"computed","paper":{"title":"Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Ahmed Asiri, Chenhan Zhang, Feng Wu, Jianhuan Huang, Luoyu Chen, Shui Yu, Weiqi Wang, Zhiyi Tian","submitted_at":"2026-05-23T12:07:38Z","abstract_excerpt":"Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle the failure on unseen jailbreaks problem, base on unsupervised latent direction discovery. We propose a "},"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":"2605.24535","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-05-23T12:07:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"eb0f5fef536912ba3776bebea1b24550482e440524b6425bc595cfbe160c159a","abstract_canon_sha256":"5b1b9485f4e69b8090e42113a3b53f45d0b3341465095b5ddf6cbfc52f455521"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:45.011072Z","signature_b64":"FVPL+gtqIkBW6fu8fqIRxpdaiq0kl7fIoC38dlFXJ1loYoAYR8STTxWB+4/6RtCe2ZToQ2+LhuojqhvktO0/BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc93bb9f2ad65f2a65f11b8e4e47e03042df5ceae1509353b26edf3f142a2f58","last_reissued_at":"2026-05-26T01:03:45.010379Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:45.010379Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Ahmed Asiri, Chenhan Zhang, Feng Wu, Jianhuan Huang, Luoyu Chen, Shui Yu, Weiqi Wang, Zhiyi Tian","submitted_at":"2026-05-23T12:07:38Z","abstract_excerpt":"Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle the failure on unseen jailbreaks problem, base on unsupervised latent direction discovery. We propose a "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.24535","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/2605.24535/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":"2605.24535","created_at":"2026-05-26T01:03:45.010504+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.24535v1","created_at":"2026-05-26T01:03:45.010504+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.24535","created_at":"2026-05-26T01:03:45.010504+00:00"},{"alias_kind":"pith_short_12","alias_value":"XSJ3XHZK2ZPS","created_at":"2026-05-26T01:03:45.010504+00:00"},{"alias_kind":"pith_short_16","alias_value":"XSJ3XHZK2ZPSUZPR","created_at":"2026-05-26T01:03:45.010504+00:00"},{"alias_kind":"pith_short_8","alias_value":"XSJ3XHZK","created_at":"2026-05-26T01:03:45.010504+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/XSJ3XHZK2ZPSUZPRDOHE4R7AGB","json":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB.json","graph_json":"https://pith.science/api/pith-number/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/graph.json","events_json":"https://pith.science/api/pith-number/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/events.json","paper":"https://pith.science/paper/XSJ3XHZK"},"agent_actions":{"view_html":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB","download_json":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB.json","view_paper":"https://pith.science/paper/XSJ3XHZK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.24535&json=true","fetch_graph":"https://pith.science/api/pith-number/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/graph.json","fetch_events":"https://pith.science/api/pith-number/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/action/storage_attestation","attest_author":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/action/author_attestation","sign_citation":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/action/citation_signature","submit_replication":"https://pith.science/pith/XSJ3XHZK2ZPSUZPRDOHE4R7AGB/action/replication_record"}},"created_at":"2026-05-26T01:03:45.010504+00:00","updated_at":"2026-05-26T01:03:45.010504+00:00"}