{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:2ZC2PTGBVJHY5D26WFTCDW7DUS","short_pith_number":"pith:2ZC2PTGB","canonical_record":{"source":{"id":"2605.05938","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-07T09:46:33Z","cross_cats_sorted":[],"title_canon_sha256":"e5343e351402786483aede04add4e0a024c2d9aac3f00001cbaae360d0aa5aaf","abstract_canon_sha256":"d3419734661b1986e2835a78506251101314a325e748ab406c5bb24715054cb1"},"schema_version":"1.0"},"canonical_sha256":"d645a7ccc1aa4f8e8f5eb16621dbe3a480eac31cba8b7a5744025a504139fdab","source":{"kind":"arxiv","id":"2605.05938","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.05938","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"arxiv_version","alias_value":"2605.05938v2","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.05938","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_12","alias_value":"2ZC2PTGBVJHY","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_16","alias_value":"2ZC2PTGBVJHY5D26","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_8","alias_value":"2ZC2PTGB","created_at":"2026-08-04T02:00:26Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:2ZC2PTGBVJHY5D26WFTCDW7DUS","target":"record","payload":{"canonical_record":{"source":{"id":"2605.05938","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-07T09:46:33Z","cross_cats_sorted":[],"title_canon_sha256":"e5343e351402786483aede04add4e0a024c2d9aac3f00001cbaae360d0aa5aaf","abstract_canon_sha256":"d3419734661b1986e2835a78506251101314a325e748ab406c5bb24715054cb1"},"schema_version":"1.0"},"canonical_sha256":"d645a7ccc1aa4f8e8f5eb16621dbe3a480eac31cba8b7a5744025a504139fdab","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T02:00:26.297051Z","signature_b64":"c5mtQbqhNhh3O2Wbo9q0wRhQfpmrb1/Aoeu/HR/YNYKj6WFwzMSi5u7ai21W/XHyTg/nGAAn0qmQK4iXBgROCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d645a7ccc1aa4f8e8f5eb16621dbe3a480eac31cba8b7a5744025a504139fdab","last_reissued_at":"2026-08-04T02:00:26.295343Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T02:00:26.295343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.05938","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-08-04T02:00:26Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"W6sw9QfTtuf85vkl+fkdbC8vFsIpUyHNQD5VFruZEoOyNcV2v5fzy3jsRZYnBBQ9j/mmSxhGKp20YAb8sLphAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T22:57:35.225887Z"},"content_sha256":"fa8c8db59c1b019166068cb4714a74fcb74ce41cd306748ef59f1a2f12a9dac5","schema_version":"1.0","event_id":"sha256:fa8c8db59c1b019166068cb4714a74fcb74ce41cd306748ef59f1a2f12a9dac5"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:2ZC2PTGBVJHY5D26WFTCDW7DUS","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"ICU-Bench:Benchmarking Continual Unlearning in Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Guangyu He, Haichang Gao, Junkai Zhang, Wenjie Mei, Yuhang Wang, Zhenxing Niu","submitted_at":"2026-05-07T09:46:33Z","abstract_excerpt":"Privacy deletion requests often arrive sequentially, creating a continual unlearning challenge for deployed multimodal large language models (MLLMs). However, existing benchmarks mainly focus on static or short-sequence settings, offering limited support for evaluating continual privacy deletion on privacy-critical documents. To bridge this gap, we introduce ICU-Bench, an Identity-centric Continual Unlearning benchmark (pronounced ``I see you'') for privacy-critical multimodal documents. ICU-Bench contains 1,000 synthetic privacy-sensitive profiles from medical reports and labor contracts, com"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 100 forget tasks and chosen document domains in ICU-Bench accurately represent realistic continual privacy deletion requests faced by MLLMs in deployment.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ICU-Bench is a new continual unlearning benchmark for MLLMs using 1000 privacy profiles, 9500 images, and 100 forget tasks, showing existing methods fail to balance forgetting, utility, and scalability.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7e63a9fb4e99f2e72cb242af09550bf4478336ae9723b7634c49b5d9ebc48b83"},"source":{"id":"2605.05938","kind":"arxiv","version":2},"verdict":{"id":"4ccfc718-b79b-4680-98f2-29b8a7b7d408","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T10:49:39.543920Z","strongest_claim":"existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences.","one_line_summary":"ICU-Bench is a new continual unlearning benchmark for MLLMs using 1000 privacy profiles, 9500 images, and 100 forget tasks, showing existing methods fail to balance forgetting, utility, and scalability.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 100 forget tasks and chosen document domains in ICU-Bench accurately represent realistic continual privacy deletion requests faced by MLLMs in deployment.","pith_extraction_headline":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.05938/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T13:22:04.405813Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T08:39:58.514920Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T19:31:19.370078Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T13:06:37.194068Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"005691930b75f673dc55df8d293dee8fcb31a9947395b78d948e0a222bef66ca"},"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":"4ccfc718-b79b-4680-98f2-29b8a7b7d408"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-08-04T02:00:26Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3XD+g2WmFwMJx029PMLGhklzke3v41ojwEFXzrerOdMaPlM3p3gKDsR40Q3n+yQdY00frLnb+HZylaNt1/JKAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-06T22:57:35.226618Z"},"content_sha256":"5ec79e0a9352617d610b81d19e1f116cfd84fbcff8598fafd2cdccc991de9392","schema_version":"1.0","event_id":"sha256:5ec79e0a9352617d610b81d19e1f116cfd84fbcff8598fafd2cdccc991de9392"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/bundle.json","state_url":"https://pith.science/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/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":"2026-08-06T22:57:35Z","links":{"resolver":"https://pith.science/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS","bundle":"https://pith.science/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/bundle.json","state":"https://pith.science/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/state.json","well_known_bundle":"https://pith.science/.well-known/pith/2ZC2PTGBVJHY5D26WFTCDW7DUS/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:2ZC2PTGBVJHY5D26WFTCDW7DUS","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":"d3419734661b1986e2835a78506251101314a325e748ab406c5bb24715054cb1","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-07T09:46:33Z","title_canon_sha256":"e5343e351402786483aede04add4e0a024c2d9aac3f00001cbaae360d0aa5aaf"},"schema_version":"1.0","source":{"id":"2605.05938","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.05938","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"arxiv_version","alias_value":"2605.05938v2","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.05938","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_12","alias_value":"2ZC2PTGBVJHY","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_16","alias_value":"2ZC2PTGBVJHY5D26","created_at":"2026-08-04T02:00:26Z"},{"alias_kind":"pith_short_8","alias_value":"2ZC2PTGB","created_at":"2026-08-04T02:00:26Z"}],"graph_snapshots":[{"event_id":"sha256:5ec79e0a9352617d610b81d19e1f116cfd84fbcff8598fafd2cdccc991de9392","target":"graph","created_at":"2026-08-04T02:00:26Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The 100 forget tasks and chosen document domains in ICU-Bench accurately represent realistic continual privacy deletion requests faced by MLLMs in deployment."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"ICU-Bench is a new continual unlearning benchmark for MLLMs using 1000 privacy profiles, 9500 images, and 100 forget tasks, showing existing methods fail to balance forgetting, utility, and scalability."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests."}],"snapshot_sha256":"7e63a9fb4e99f2e72cb242af09550bf4478336ae9723b7634c49b5d9ebc48b83"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T13:22:04.405813Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T08:39:58.514920Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T19:31:19.370078Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T13:06:37.194068Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.05938/integrity.json","findings":[],"snapshot_sha256":"005691930b75f673dc55df8d293dee8fcb31a9947395b78d948e0a222bef66ca","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Privacy deletion requests often arrive sequentially, creating a continual unlearning challenge for deployed multimodal large language models (MLLMs). However, existing benchmarks mainly focus on static or short-sequence settings, offering limited support for evaluating continual privacy deletion on privacy-critical documents. To bridge this gap, we introduce ICU-Bench, an Identity-centric Continual Unlearning benchmark (pronounced ``I see you'') for privacy-critical multimodal documents. ICU-Bench contains 1,000 synthetic privacy-sensitive profiles from medical reports and labor contracts, com","authors_text":"Guangyu He, Haichang Gao, Junkai Zhang, Wenjie Mei, Yuhang Wang, Zhenxing Niu","cross_cats":[],"headline":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests.","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-07T09:46:33Z","title":"ICU-Bench:Benchmarking Continual Unlearning in Multimodal Large Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.05938","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-08T10:49:39.543920Z","id":"4ccfc718-b79b-4680-98f2-29b8a7b7d408","model_set":{"reader":"grok-4.3"},"one_line_summary":"ICU-Bench is a new continual unlearning benchmark for MLLMs using 1000 privacy profiles, 9500 images, and 100 forget tasks, showing existing methods fail to balance forgetting, utility, and scalability.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Existing unlearning methods for multimodal models fail to scale across repeated privacy deletion requests.","strongest_claim":"existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences.","weakest_assumption":"The 100 forget tasks and chosen document domains in ICU-Bench accurately represent realistic continual privacy deletion requests faced by MLLMs in deployment."}},"verdict_id":"4ccfc718-b79b-4680-98f2-29b8a7b7d408"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:fa8c8db59c1b019166068cb4714a74fcb74ce41cd306748ef59f1a2f12a9dac5","target":"record","created_at":"2026-08-04T02:00:26Z","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":"d3419734661b1986e2835a78506251101314a325e748ab406c5bb24715054cb1","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-07T09:46:33Z","title_canon_sha256":"e5343e351402786483aede04add4e0a024c2d9aac3f00001cbaae360d0aa5aaf"},"schema_version":"1.0","source":{"id":"2605.05938","kind":"arxiv","version":2}},"canonical_sha256":"d645a7ccc1aa4f8e8f5eb16621dbe3a480eac31cba8b7a5744025a504139fdab","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d645a7ccc1aa4f8e8f5eb16621dbe3a480eac31cba8b7a5744025a504139fdab","first_computed_at":"2026-08-04T02:00:26.295343Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-08-04T02:00:26.295343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"c5mtQbqhNhh3O2Wbo9q0wRhQfpmrb1/Aoeu/HR/YNYKj6WFwzMSi5u7ai21W/XHyTg/nGAAn0qmQK4iXBgROCw==","signature_status":"signed_v1","signed_at":"2026-08-04T02:00:26.297051Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.05938","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:fa8c8db59c1b019166068cb4714a74fcb74ce41cd306748ef59f1a2f12a9dac5","sha256:5ec79e0a9352617d610b81d19e1f116cfd84fbcff8598fafd2cdccc991de9392"],"state_sha256":"4480ec952d9247fb437a8d847c93cd7e1fb1a983e2f10cae8e79d0dad1e3fab4"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"iGL4FiJmRzgrXpsF+9f8E62PgfR4dNFnoGv36aoxszi9C5mRf1zKlzhO57ZEOZtoXkCr2H/y21tNKHuVIn6GAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-06T22:57:35.231070Z","bundle_sha256":"51fc3394e4a990276bac3e394d5f47e83bc614bc08e2ebfb87559ad8fcced688"}}