{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RKBWFDVSD5Z3AAO6GXBFJLM6CL","short_pith_number":"pith:RKBWFDVS","schema_version":"1.0","canonical_sha256":"8a83628eb21f73b001de35c254ad9e12d13573ca7593b99a4be64abee1194d03","source":{"kind":"arxiv","id":"2502.11184","version":2},"attestation_state":"computed","paper":{"title":"Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.MM"],"primary_cat":"cs.CL","authors_text":"Jen-tse Huang, Kuiyi Gao, Pinjia He, Shuai Wang, Wenxuan Wang, Xiaoyuan Liu, Youliang Yuan, Zhaopeng Tu","submitted_at":"2025-02-16T16:12:40Z","abstract_excerpt":"Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. However, ensuring the safety of these models remains a significant challenge, particularly in accurately identifying whether multimodal content is safe or unsafe-a capability we term safety awareness. In this paper, we introduce MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios with 1500 carefully curated image-prompt pairs. MMSafeAware includes both unsafe and "},"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":"2502.11184","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-16T16:12:40Z","cross_cats_sorted":["cs.AI","cs.CV","cs.MM"],"title_canon_sha256":"ae8c044363bf3260cc7a43801513965a83893753be32736bb4b2f063b2a8c8b1","abstract_canon_sha256":"2d013cef826dca4236db73daea47e82114b00b5a706b939786c0ebb31215690d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:53.850078Z","signature_b64":"AgbMpPgBC1ETJJD/MjKp10v60Fa+SOnRI7Cbp1ccnKkwvHTi/rq+1+laRn8boydqwjZSOJkCZwwKSItEFN6aBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8a83628eb21f73b001de35c254ad9e12d13573ca7593b99a4be64abee1194d03","last_reissued_at":"2026-07-05T11:14:53.849645Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:53.849645Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Can't See the Forest for the Trees: Benchmarking Multimodal Safety Awareness for Multimodal LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.MM"],"primary_cat":"cs.CL","authors_text":"Jen-tse Huang, Kuiyi Gao, Pinjia He, Shuai Wang, Wenxuan Wang, Xiaoyuan Liu, Youliang Yuan, Zhaopeng Tu","submitted_at":"2025-02-16T16:12:40Z","abstract_excerpt":"Multimodal Large Language Models (MLLMs) have expanded the capabilities of traditional language models by enabling interaction through both text and images. However, ensuring the safety of these models remains a significant challenge, particularly in accurately identifying whether multimodal content is safe or unsafe-a capability we term safety awareness. In this paper, we introduce MMSafeAware, the first comprehensive multimodal safety awareness benchmark designed to evaluate MLLMs across 29 safety scenarios with 1500 carefully curated image-prompt pairs. MMSafeAware includes both unsafe and "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.11184","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/2502.11184/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":"2502.11184","created_at":"2026-07-05T11:14:53.849702+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.11184v2","created_at":"2026-07-05T11:14:53.849702+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.11184","created_at":"2026-07-05T11:14:53.849702+00:00"},{"alias_kind":"pith_short_12","alias_value":"RKBWFDVSD5Z3","created_at":"2026-07-05T11:14:53.849702+00:00"},{"alias_kind":"pith_short_16","alias_value":"RKBWFDVSD5Z3AAO6","created_at":"2026-07-05T11:14:53.849702+00:00"},{"alias_kind":"pith_short_8","alias_value":"RKBWFDVS","created_at":"2026-07-05T11:14:53.849702+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07335","citing_title":"Defending Jailbreak Attacks on Large Language Models via Manifold Trajectory Kinetics","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11716","citing_title":"SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL","json":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL.json","graph_json":"https://pith.science/api/pith-number/RKBWFDVSD5Z3AAO6GXBFJLM6CL/graph.json","events_json":"https://pith.science/api/pith-number/RKBWFDVSD5Z3AAO6GXBFJLM6CL/events.json","paper":"https://pith.science/paper/RKBWFDVS"},"agent_actions":{"view_html":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL","download_json":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL.json","view_paper":"https://pith.science/paper/RKBWFDVS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.11184&json=true","fetch_graph":"https://pith.science/api/pith-number/RKBWFDVSD5Z3AAO6GXBFJLM6CL/graph.json","fetch_events":"https://pith.science/api/pith-number/RKBWFDVSD5Z3AAO6GXBFJLM6CL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL/action/storage_attestation","attest_author":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL/action/author_attestation","sign_citation":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL/action/citation_signature","submit_replication":"https://pith.science/pith/RKBWFDVSD5Z3AAO6GXBFJLM6CL/action/replication_record"}},"created_at":"2026-07-05T11:14:53.849702+00:00","updated_at":"2026-07-05T11:14:53.849702+00:00"}