{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:AX5VZLE7WIKRFR3BXXPZ7FPAZS","short_pith_number":"pith:AX5VZLE7","canonical_record":{"source":{"id":"2608.10513","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-11T05:37:59Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cd193e7ab80dd938380e4f9d1b9df0e24e8df57dcdeeedf17108851a62db0620","abstract_canon_sha256":"ced1769daa9aa4318a0b1f90119b42eb06c683c4d774fde0a06383eafae039c3"},"schema_version":"1.0"},"canonical_sha256":"05fb5cac9fb21512c761bddf9f95e0cc92492d3d377bed11d49018efb8c70d51","source":{"kind":"arxiv","id":"2608.10513","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2608.10513","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"arxiv_version","alias_value":"2608.10513v1","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.10513","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_12","alias_value":"AX5VZLE7WIKR","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_16","alias_value":"AX5VZLE7WIKRFR3B","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_8","alias_value":"AX5VZLE7","created_at":"2026-08-12T01:22:38Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:AX5VZLE7WIKRFR3BXXPZ7FPAZS","target":"record","payload":{"canonical_record":{"source":{"id":"2608.10513","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-11T05:37:59Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cd193e7ab80dd938380e4f9d1b9df0e24e8df57dcdeeedf17108851a62db0620","abstract_canon_sha256":"ced1769daa9aa4318a0b1f90119b42eb06c683c4d774fde0a06383eafae039c3"},"schema_version":"1.0"},"canonical_sha256":"05fb5cac9fb21512c761bddf9f95e0cc92492d3d377bed11d49018efb8c70d51","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-12T01:22:38.783383Z","signature_b64":"sCD1G1MprMCGpimecTJ2DvBJax6KG+3t8X5KqG79A2ACRgkXqLEdlOQp+8bM4hP2MHAABfCgesgFlw9bBUrvDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"05fb5cac9fb21512c761bddf9f95e0cc92492d3d377bed11d49018efb8c70d51","last_reissued_at":"2026-08-12T01:22:38.781813Z","signature_status":"signed_v1","first_computed_at":"2026-08-12T01:22:38.781813Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2608.10513","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-08-12T01:22:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"htguar6wuqynBEuWFkn62PL98QuFEkt3o5XdHR/PfRpqwtWZTaAnw86ff/Q4Fhu57SXEWpK3DUdEhN0PFatxBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-15T21:51:29.037722Z"},"content_sha256":"e7a33581b27c58af005c48141da2bed7047b9675dcf867c8e3c9518fb83be80b","schema_version":"1.0","event_id":"sha256:e7a33581b27c58af005c48141da2bed7047b9675dcf867c8e3c9518fb83be80b"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:AX5VZLE7WIKRFR3BXXPZ7FPAZS","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SafeCap: Improving LVLM Safety with Image Captioning Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Caoyuan Ma, Ji Zhao, Lingxi Min, Shilei Zhao, Shuai Dong, Taiqiang Wu, Tian Gu, Weichu Xie, Wenpu Liu, Wenqi Shao, Wenzheng Chang, Yinqiang Zheng, Yongfu Zhu, Yuqi Xu, Ziyue Wang","submitted_at":"2026-08-11T05:37:59Z","abstract_excerpt":"Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones. We propose SafeCap, a reinforcement-learning framework that aligns LVLMs through learned self-captioning. SafeCap trains a policy model to first generate a safety-relevant image caption and then produce a final answer; the caption is further optimized by whether it enables a frozen LLM to reach a safety-aligned decision. This caption-mediated objective encourages the policy to expose visual cues relevant to safe response gene"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.10513","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/2608.10513/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-08-12T01:22:38Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fYWpUYypz8pyeeM/+B2RpczfEfH8UcN/KFAwzC0Bl7+78tEpe8rSMaXifujzCt774HComhD6vbXFIctKT4SsAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-15T21:51:29.038289Z"},"content_sha256":"c9b987dbb6eca5560797fe63f4886dfb9b7c1e4eb774ba9f5be83d27fd75d7fe","schema_version":"1.0","event_id":"sha256:c9b987dbb6eca5560797fe63f4886dfb9b7c1e4eb774ba9f5be83d27fd75d7fe"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/bundle.json","state_url":"https://pith.science/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-15T21:51:29Z","links":{"resolver":"https://pith.science/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS","bundle":"https://pith.science/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/bundle.json","state":"https://pith.science/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/state.json","well_known_bundle":"https://pith.science/.well-known/pith/AX5VZLE7WIKRFR3BXXPZ7FPAZS/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:AX5VZLE7WIKRFR3BXXPZ7FPAZS","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"ced1769daa9aa4318a0b1f90119b42eb06c683c4d774fde0a06383eafae039c3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-11T05:37:59Z","title_canon_sha256":"cd193e7ab80dd938380e4f9d1b9df0e24e8df57dcdeeedf17108851a62db0620"},"schema_version":"1.0","source":{"id":"2608.10513","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2608.10513","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"arxiv_version","alias_value":"2608.10513v1","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.10513","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_12","alias_value":"AX5VZLE7WIKR","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_16","alias_value":"AX5VZLE7WIKRFR3B","created_at":"2026-08-12T01:22:38Z"},{"alias_kind":"pith_short_8","alias_value":"AX5VZLE7","created_at":"2026-08-12T01:22:38Z"}],"graph_snapshots":[{"event_id":"sha256:c9b987dbb6eca5560797fe63f4886dfb9b7c1e4eb774ba9f5be83d27fd75d7fe","target":"graph","created_at":"2026-08-12T01:22:38Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2608.10513/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Large vision-language models (LVLMs) remain vulnerable to jailbreak attacks that exploit visual inputs to bypass safety alignment inherited from their language backbones. We propose SafeCap, a reinforcement-learning framework that aligns LVLMs through learned self-captioning. SafeCap trains a policy model to first generate a safety-relevant image caption and then produce a final answer; the caption is further optimized by whether it enables a frozen LLM to reach a safety-aligned decision. This caption-mediated objective encourages the policy to expose visual cues relevant to safe response gene","authors_text":"Caoyuan Ma, Ji Zhao, Lingxi Min, Shilei Zhao, Shuai Dong, Taiqiang Wu, Tian Gu, Weichu Xie, Wenpu Liu, Wenqi Shao, Wenzheng Chang, Yinqiang Zheng, Yongfu Zhu, Yuqi Xu, Ziyue Wang","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-11T05:37:59Z","title":"SafeCap: Improving LVLM Safety with Image Captioning Reinforcement Learning"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.10513","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:e7a33581b27c58af005c48141da2bed7047b9675dcf867c8e3c9518fb83be80b","target":"record","created_at":"2026-08-12T01:22:38Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"ced1769daa9aa4318a0b1f90119b42eb06c683c4d774fde0a06383eafae039c3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-08-11T05:37:59Z","title_canon_sha256":"cd193e7ab80dd938380e4f9d1b9df0e24e8df57dcdeeedf17108851a62db0620"},"schema_version":"1.0","source":{"id":"2608.10513","kind":"arxiv","version":1}},"canonical_sha256":"05fb5cac9fb21512c761bddf9f95e0cc92492d3d377bed11d49018efb8c70d51","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"05fb5cac9fb21512c761bddf9f95e0cc92492d3d377bed11d49018efb8c70d51","first_computed_at":"2026-08-12T01:22:38.781813Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-08-12T01:22:38.781813Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"sCD1G1MprMCGpimecTJ2DvBJax6KG+3t8X5KqG79A2ACRgkXqLEdlOQp+8bM4hP2MHAABfCgesgFlw9bBUrvDw==","signature_status":"signed_v1","signed_at":"2026-08-12T01:22:38.783383Z","signed_message":"canonical_sha256_bytes"},"source_id":"2608.10513","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:e7a33581b27c58af005c48141da2bed7047b9675dcf867c8e3c9518fb83be80b","sha256:c9b987dbb6eca5560797fe63f4886dfb9b7c1e4eb774ba9f5be83d27fd75d7fe"],"state_sha256":"a8d5ffa9cfd83eb3c6c55eec1d6e0557094682f62fe8e6b27ca4cf0e1de30d02"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"FcdbZAq/ggQve/dAS0LRgJPQpUFTK4ZchxV7GR256UY+EKSvq2cRbylzO0C8LvGqbzTtMmgjXVhoCx8665ZeAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-15T21:51:29.042137Z","bundle_sha256":"e058dcd171aa8860554066a0308bab60a52e9dccba6755ede3e111837b3f53be"}}