{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:RPJBTYC2XU2WDEILHM22GV23MC","merge_version":"pith-open-graph-merge-v1","event_count":3,"valid_event_count":3,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a9585556d4eac7ba3f31bda5eab6158230f55cba1bd4d8ebcc5d28d5b8c7fd4c","cross_cats_sorted":["cs.AI","cs.CR","cs.CY"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T22:07:22Z","title_canon_sha256":"2793d649b92807e524b737580e81e2ac74b9eb9ed82aaa45ac210936b7a05169"},"schema_version":"1.0","source":{"id":"2605.14152","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14152","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14152v1","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14152","created_at":"2026-05-17T23:39:11Z"},{"alias_kind":"pith_short_12","alias_value":"RPJBTYC2XU2W","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"RPJBTYC2XU2WDEIL","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"RPJBTYC2","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:77305d9035c51ba42394939062ad4e51d10a1555049263be842eef3e632d908b","target":"graph","created_at":"2026-05-17T23:39:11Z","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":"Across a dual-track set of frontier and Korean-optimized models, we find a consistent suppression effect in Korean variants and substantial model-to-model variation in how geopolitical grounding interacts with language. In many models, Korean grounding mitigates the Korean language-driven suppression -- with no model showing significant amplification in the other direction."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the expert-crafted binary rubrics and LLM-as-a-judge panels produce stable, unbiased safety scores that correctly separate language effects from geopolitical effects without introducing their own cultural or linguistic biases."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"ROK-FORTRESS shows Korean-language prompts increase LLM safety suppression compared with English, while Korean geopolitical grounding often reduces that suppression, indicating translation-only evaluations miss language-context interactions."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Korean language increases suppression of responses to security prompts in LLMs, while Korean geopolitical context often reduces it."}],"snapshot_sha256":"9bbd6a5ed12b08890fe145babf02fd296af5e0a55df3895f40f2654c99348385"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Safety evaluations for large language models (LLMs) increasingly target high-stakes National Security and Public Safety (NSPS) risks, yet multilingual safety is typically assessed through translation-only benchmarks that preserve the underlying scenario, and empirical evidence of how language and geopolitical context interact remains limited to a narrow set of language pairs. We introduce \\emph{ROK-FORTRESS} https://huggingface.co/datasets/ScaleAI/ROK-FORTRESS_public, a bilingual, culturally adversarial NSPS benchmark that uses the English--Korean language pair and U.S.--ROK geopolitical axis ","authors_text":"Bert Herring, Christina Q Knight, Drew Rein, Evi Fuelle, Jiyeon Cho, Jiyeon Joo, Jonathan Nguyen, Joseph Brandifino, Kaustubh Deshpande, Kyungho Song, Max Fenkell, Michael S. Lee, Minn Seok Choi, Udari Madhushani Sehwag, Yash Maurya, Yeongkyun Jang","cross_cats":["cs.AI","cs.CR","cs.CY"],"headline":"Korean language increases suppression of responses to security prompts in LLMs, while Korean geopolitical context often reduces it.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T22:07:22Z","title":"ROK-FORTRESS: Measuring the Effect of Geopolitical Transcreation for National Security and Public Safety"},"references":{"count":67,"internal_anchors":0,"resolved_work":67,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"The multilingual alignment prism: Aligning global and local prefer- ences to reduce harm","work_id":"58ab5626-7190-4afc-9ef0-f99f768ea4aa","year":null},{"cited_arxiv_id":"","doi":"10.48550/arxiv.2406.18682","is_internal_anchor":false,"ref_index":2,"title":"The multilingual alignment prism: Aligning global and local prefer- ences to reduce harm","work_id":"58ab5626-7190-4afc-9ef0-f99f768ea4aa","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"M. Banko, A. Vella, B. Ray, E. Strubell, H. Wallach, and Y. Elazar. A unified typology of harmful content.Proceedings of the First Workshop on Online Abuse and Harms (ALW), pages 1–15, 2020. URL https","work_id":"d13fabb4-2606-4f66-a1f1-5b94d46001e1","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"N. Bostrom. Information hazards: A typology of potential harms from knowledge.Review of Contemporary Philosophy, 10:44–79, 2011. URLhttps://nickbostrom.com/information-hazards.pdf. Accessed 2026","work_id":"449b92ff-4e49-4df4-9c5c-88a3b14f1859","year":2011},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Multilingual jailbreak challenges in large language models.arXiv preprint arXiv:2310.06474","work_id":"97d238ad-7e44-4a09-90f1-ca400f5296e3","year":2023}],"snapshot_sha256":"846ba1f8233b279f3df8d223c16913db45d6b63e1a9f4be010b11d8522b56999"},"source":{"id":"2605.14152","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T04:52:27.711968Z","id":"9bdbeeae-a591-4367-aa98-e50616c8c0c2","model_set":{"reader":"grok-4.3"},"one_line_summary":"ROK-FORTRESS shows Korean-language prompts increase LLM safety suppression compared with English, while Korean geopolitical grounding often reduces that suppression, indicating translation-only evaluations miss language-context interactions.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Korean language increases suppression of responses to security prompts in LLMs, while Korean geopolitical context often reduces it.","strongest_claim":"Across a dual-track set of frontier and Korean-optimized models, we find a consistent suppression effect in Korean variants and substantial model-to-model variation in how geopolitical grounding interacts with language. In many models, Korean grounding mitigates the Korean language-driven suppression -- with no model showing significant amplification in the other direction.","weakest_assumption":"That the expert-crafted binary rubrics and LLM-as-a-judge panels produce stable, unbiased safety scores that correctly separate language effects from geopolitical effects without introducing their own cultural or linguistic biases."}},"verdict_id":"9bdbeeae-a591-4367-aa98-e50616c8c0c2"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6bfdde247a1efb169edc275b625e06ce8ec9a445e3abc6201b827bd3384201fa","target":"record","created_at":"2026-05-17T23:39:11Z","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":"a9585556d4eac7ba3f31bda5eab6158230f55cba1bd4d8ebcc5d28d5b8c7fd4c","cross_cats_sorted":["cs.AI","cs.CR","cs.CY"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T22:07:22Z","title_canon_sha256":"2793d649b92807e524b737580e81e2ac74b9eb9ed82aaa45ac210936b7a05169"},"schema_version":"1.0","source":{"id":"2605.14152","kind":"arxiv","version":1}},"canonical_sha256":"8bd219e05abd3561910b3b35a3575b60ab30fc743bca3ef4d610a497ef5a7dd0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"8bd219e05abd3561910b3b35a3575b60ab30fc743bca3ef4d610a497ef5a7dd0","first_computed_at":"2026-05-17T23:39:11.573817Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:11.573817Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HUyj2WGR0fXRG4dlU52XpF1EunFk68JTrl0/hTiYtYFRkw9Uj9EShjgzyR3LM7qfHPEu6Kf/SQ9HQenVFbAXDQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:11.574366Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14152","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6bfdde247a1efb169edc275b625e06ce8ec9a445e3abc6201b827bd3384201fa","sha256:77305d9035c51ba42394939062ad4e51d10a1555049263be842eef3e632d908b","sha256:01bf933a2ebc60f73926ba4bbb7dedaef6089c7a88240a67b79ec288c42594ae"],"state_sha256":"b48de0886d4088ed06228851f4789a057e3cb3b6cb30a9d35b65743e86c1f6fb"}