{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:75LRLKFF3R4NVGL42HF5VSREXP","short_pith_number":"pith:75LRLKFF","schema_version":"1.0","canonical_sha256":"ff5715a8a5dc78da997cd1cbdaca24bbdeee3c20550d8a315b898c225757c5bb","source":{"kind":"arxiv","id":"2505.18979","version":2},"attestation_state":"computed","paper":{"title":"Dynamic Optimization and Safety Indicator Injection for Jailbreaking Text-to-Image Models with Multimodal Safety Filters","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hao Lin, Ke Xu, Tanfeng Sun, Xinghao Jiang, Zixuan Chen","submitted_at":"2025-05-25T05:13:06Z","abstract_excerpt":"Text-to-image (T2I) models can generate not-safe-for-work (NSFW) content, motivating multi-stage safety pipelines with both text and image filters. Newer LLM-based filters detect latent intent beyond keywords, making token-level perturbation attacks unreliable. Our evaluation further shows that existing jailbreak methods exhibit a sharp trade-off between filter evasion and semantic fidelity, while also requiring excessive queries to succeed. We introduce \\textbf{OptJail}, an automated jailbreak framework that combines dynamic prompt optimization with multimodal feedback. It consists of two key"},"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":"2505.18979","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-25T05:13:06Z","cross_cats_sorted":[],"title_canon_sha256":"18078343b5c3e3e0fab0b596bab7cd5bb42138b3dda7b36f51cee0949190d2d1","abstract_canon_sha256":"65d85b9f81f1d97b6df2a1a0636a12fa5e3d1a608d636e6a1255529f4802cf0a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:12.804292Z","signature_b64":"wD+6N5UaqWsAv38J5201COtXODGGyhY0THNr1vB7BGJydUgoyzdsGiP+R8G4WPP4b8Iy1OZlCFFyY5tXDRmbDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ff5715a8a5dc78da997cd1cbdaca24bbdeee3c20550d8a315b898c225757c5bb","last_reissued_at":"2026-05-26T01:03:12.803688Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:12.803688Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dynamic Optimization and Safety Indicator Injection for Jailbreaking Text-to-Image Models with Multimodal Safety Filters","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Hao Lin, Ke Xu, Tanfeng Sun, Xinghao Jiang, Zixuan Chen","submitted_at":"2025-05-25T05:13:06Z","abstract_excerpt":"Text-to-image (T2I) models can generate not-safe-for-work (NSFW) content, motivating multi-stage safety pipelines with both text and image filters. Newer LLM-based filters detect latent intent beyond keywords, making token-level perturbation attacks unreliable. Our evaluation further shows that existing jailbreak methods exhibit a sharp trade-off between filter evasion and semantic fidelity, while also requiring excessive queries to succeed. We introduce \\textbf{OptJail}, an automated jailbreak framework that combines dynamic prompt optimization with multimodal feedback. It consists of two key"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18979","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/2505.18979/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":"2505.18979","created_at":"2026-05-26T01:03:12.803764+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18979v2","created_at":"2026-05-26T01:03:12.803764+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18979","created_at":"2026-05-26T01:03:12.803764+00:00"},{"alias_kind":"pith_short_12","alias_value":"75LRLKFF3R4N","created_at":"2026-05-26T01:03:12.803764+00:00"},{"alias_kind":"pith_short_16","alias_value":"75LRLKFF3R4NVGL4","created_at":"2026-05-26T01:03:12.803764+00:00"},{"alias_kind":"pith_short_8","alias_value":"75LRLKFF","created_at":"2026-05-26T01:03:12.803764+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.18150","citing_title":"Whispers in the Noise: Surrogate-Guided Concept Awakening via a Multi-Agent Framework","ref_index":2,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP","json":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP.json","graph_json":"https://pith.science/api/pith-number/75LRLKFF3R4NVGL42HF5VSREXP/graph.json","events_json":"https://pith.science/api/pith-number/75LRLKFF3R4NVGL42HF5VSREXP/events.json","paper":"https://pith.science/paper/75LRLKFF"},"agent_actions":{"view_html":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP","download_json":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP.json","view_paper":"https://pith.science/paper/75LRLKFF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18979&json=true","fetch_graph":"https://pith.science/api/pith-number/75LRLKFF3R4NVGL42HF5VSREXP/graph.json","fetch_events":"https://pith.science/api/pith-number/75LRLKFF3R4NVGL42HF5VSREXP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP/action/storage_attestation","attest_author":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP/action/author_attestation","sign_citation":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP/action/citation_signature","submit_replication":"https://pith.science/pith/75LRLKFF3R4NVGL42HF5VSREXP/action/replication_record"}},"created_at":"2026-05-26T01:03:12.803764+00:00","updated_at":"2026-05-26T01:03:12.803764+00:00"}