{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MO2AEVWG7VLD3KHS6MO6QSOJ65","short_pith_number":"pith:MO2AEVWG","schema_version":"1.0","canonical_sha256":"63b40256c6fd563da8f2f31de849c9f76852ff83a044535d7e688eb16082bb85","source":{"kind":"arxiv","id":"2505.18556","version":2},"attestation_state":"computed","paper":{"title":"Exploring the Vulnerability of the Content Moderation Guardrail in Large Language Models via Intent Manipulation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gaby G. Dagher, Haibo Jin, Haohan Wang, Jun Zhuang, Wenbin Zhang, Ye Zhang, Zhengjian Kang","submitted_at":"2025-05-24T06:47:32Z","abstract_excerpt":"Intent detection, a core component of natural language understanding, has considerably evolved as a crucial mechanism in safeguarding large language models (LLMs). While prior work has applied intent detection to enhance LLMs' moderation guardrails, showing a significant success against content-level jailbreaks, the robustness of these intent-aware guardrails under malicious manipulations remains under-explored. In this work, we investigate the vulnerability of intent-aware guardrails and demonstrate that LLMs exhibit implicit intent detection capabilities. We propose a two-stage intent-based "},"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":"2505.18556","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-24T06:47:32Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"390b8eb03853a5416533e3651c89ecb4724480dc13db7d9a7d43c68a8cb494eb","abstract_canon_sha256":"b8ff1f647133c988c95887e317daa3ecda5086041fb628ea86fe119dfec57499"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:58:29.746210Z","signature_b64":"elUeJUE28k1Ts9ZAPZArWMLbj96LU216Kk1APLsWrcvV/HtYe/zAf10Mgu0i4Pw4MpxVZZ8WzBx4IoGHGA9cDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63b40256c6fd563da8f2f31de849c9f76852ff83a044535d7e688eb16082bb85","last_reissued_at":"2026-07-05T11:58:29.745694Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:58:29.745694Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploring the Vulnerability of the Content Moderation Guardrail in Large Language Models via Intent Manipulation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gaby G. Dagher, Haibo Jin, Haohan Wang, Jun Zhuang, Wenbin Zhang, Ye Zhang, Zhengjian Kang","submitted_at":"2025-05-24T06:47:32Z","abstract_excerpt":"Intent detection, a core component of natural language understanding, has considerably evolved as a crucial mechanism in safeguarding large language models (LLMs). While prior work has applied intent detection to enhance LLMs' moderation guardrails, showing a significant success against content-level jailbreaks, the robustness of these intent-aware guardrails under malicious manipulations remains under-explored. In this work, we investigate the vulnerability of intent-aware guardrails and demonstrate that LLMs exhibit implicit intent detection capabilities. We propose a two-stage intent-based "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18556","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/2505.18556/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":"2505.18556","created_at":"2026-07-05T11:58:29.745762+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18556v2","created_at":"2026-07-05T11:58:29.745762+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18556","created_at":"2026-07-05T11:58:29.745762+00:00"},{"alias_kind":"pith_short_12","alias_value":"MO2AEVWG7VLD","created_at":"2026-07-05T11:58:29.745762+00:00"},{"alias_kind":"pith_short_16","alias_value":"MO2AEVWG7VLD3KHS","created_at":"2026-07-05T11:58:29.745762+00:00"},{"alias_kind":"pith_short_8","alias_value":"MO2AEVWG","created_at":"2026-07-05T11:58:29.745762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.10546","citing_title":"Learning to Conceal Risk: Controllable Multi-turn Red Teaming for LLMs in the Financial Domain","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2601.02670","citing_title":"Break Me If You Can: Self-Jailbreaking of Aligned LLMs via Lexical Insertion Prompting","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65","json":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65.json","graph_json":"https://pith.science/api/pith-number/MO2AEVWG7VLD3KHS6MO6QSOJ65/graph.json","events_json":"https://pith.science/api/pith-number/MO2AEVWG7VLD3KHS6MO6QSOJ65/events.json","paper":"https://pith.science/paper/MO2AEVWG"},"agent_actions":{"view_html":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65","download_json":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65.json","view_paper":"https://pith.science/paper/MO2AEVWG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18556&json=true","fetch_graph":"https://pith.science/api/pith-number/MO2AEVWG7VLD3KHS6MO6QSOJ65/graph.json","fetch_events":"https://pith.science/api/pith-number/MO2AEVWG7VLD3KHS6MO6QSOJ65/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65/action/storage_attestation","attest_author":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65/action/author_attestation","sign_citation":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65/action/citation_signature","submit_replication":"https://pith.science/pith/MO2AEVWG7VLD3KHS6MO6QSOJ65/action/replication_record"}},"created_at":"2026-07-05T11:58:29.745762+00:00","updated_at":"2026-07-05T11:58:29.745762+00:00"}