{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SYSRC46WM6FPEBFZOYGWBQ53SB","short_pith_number":"pith:SYSRC46W","schema_version":"1.0","canonical_sha256":"96251173d6678af204b9760d60c3bb90565a31d4f434ebde474c03b2cb04b5e4","source":{"kind":"arxiv","id":"2508.12072","version":2},"attestation_state":"computed","paper":{"title":"Mitigating Jailbreaks with Intent-Aware LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Erik Cambria, Ranjan Satapathy, Wei Jie Yeo","submitted_at":"2025-08-16T15:03:33Z","abstract_excerpt":"Despite extensive safety-tuning, large language models (LLMs) remain vulnerable to jailbreak attacks via adversarially crafted instructions, reflecting a persistent trade-off between safety and task performance. In this work, we propose Intent-FT, a simple and lightweight fine-tuning approach that explicitly trains LLMs to infer the underlying intent of an instruction before responding. By fine-tuning on a targeted set of adversarial instructions, Intent-FT enables LLMs to generalize intent deduction to unseen attacks, thereby substantially improving their robustness. We comprehensively evalua"},"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":"2508.12072","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-08-16T15:03:33Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e71f8a572654e88e3f69edc3fe04b0300bc203354ee7016735a6b2b5254872d8","abstract_canon_sha256":"d6e3f608d67c8c5c7c500a443b232c011b43222bfa842361eeeb5018195ed006"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:57:58.939516Z","signature_b64":"x+sGmPnBPO0pA/Fzs9xIUD9bFUqLbLEX79mQh+9xTbbkaulzioQUaYpTuCmbwfuHaBsm9KV6oKP7v25MGwDVAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96251173d6678af204b9760d60c3bb90565a31d4f434ebde474c03b2cb04b5e4","last_reissued_at":"2026-07-05T11:57:58.939024Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:57:58.939024Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mitigating Jailbreaks with Intent-Aware LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Erik Cambria, Ranjan Satapathy, Wei Jie Yeo","submitted_at":"2025-08-16T15:03:33Z","abstract_excerpt":"Despite extensive safety-tuning, large language models (LLMs) remain vulnerable to jailbreak attacks via adversarially crafted instructions, reflecting a persistent trade-off between safety and task performance. In this work, we propose Intent-FT, a simple and lightweight fine-tuning approach that explicitly trains LLMs to infer the underlying intent of an instruction before responding. By fine-tuning on a targeted set of adversarial instructions, Intent-FT enables LLMs to generalize intent deduction to unseen attacks, thereby substantially improving their robustness. We comprehensively evalua"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.12072","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/2508.12072/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":"2508.12072","created_at":"2026-07-05T11:57:58.939081+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.12072v2","created_at":"2026-07-05T11:57:58.939081+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.12072","created_at":"2026-07-05T11:57:58.939081+00:00"},{"alias_kind":"pith_short_12","alias_value":"SYSRC46WM6FP","created_at":"2026-07-05T11:57:58.939081+00:00"},{"alias_kind":"pith_short_16","alias_value":"SYSRC46WM6FPEBFZ","created_at":"2026-07-05T11:57:58.939081+00:00"},{"alias_kind":"pith_short_8","alias_value":"SYSRC46W","created_at":"2026-07-05T11:57:58.939081+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06326","citing_title":"DT-Guard: Intent-Driven Reasoning-Active Training for Reasoning-Free LLM Safety Guardrail","ref_index":15,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB","json":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB.json","graph_json":"https://pith.science/api/pith-number/SYSRC46WM6FPEBFZOYGWBQ53SB/graph.json","events_json":"https://pith.science/api/pith-number/SYSRC46WM6FPEBFZOYGWBQ53SB/events.json","paper":"https://pith.science/paper/SYSRC46W"},"agent_actions":{"view_html":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB","download_json":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB.json","view_paper":"https://pith.science/paper/SYSRC46W","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.12072&json=true","fetch_graph":"https://pith.science/api/pith-number/SYSRC46WM6FPEBFZOYGWBQ53SB/graph.json","fetch_events":"https://pith.science/api/pith-number/SYSRC46WM6FPEBFZOYGWBQ53SB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB/action/storage_attestation","attest_author":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB/action/author_attestation","sign_citation":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB/action/citation_signature","submit_replication":"https://pith.science/pith/SYSRC46WM6FPEBFZOYGWBQ53SB/action/replication_record"}},"created_at":"2026-07-05T11:57:58.939081+00:00","updated_at":"2026-07-05T11:57:58.939081+00:00"}