{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:V4DWYQEJEF4FHSG6QMAP5PU4KZ","short_pith_number":"pith:V4DWYQEJ","canonical_record":{"source":{"id":"2607.01859","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-02T08:17:57Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"2b7c5c6a10e6d305504f33c41f26de4235e5f638c55e96c6ac2e2b1d6cd4a6bf","abstract_canon_sha256":"a450899c7d70ddca13647262070e5c90a1989b95c9a9a281b3deee90def965a3"},"schema_version":"1.0"},"canonical_sha256":"af076c4089217853c8de8300febe9c565079391cab672b0a12314d2b5213ed3d","source":{"kind":"arxiv","id":"2607.01859","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.01859","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"arxiv_version","alias_value":"2607.01859v1","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.01859","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_12","alias_value":"V4DWYQEJEF4F","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_16","alias_value":"V4DWYQEJEF4FHSG6","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_8","alias_value":"V4DWYQEJ","created_at":"2026-07-03T01:17:31Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:V4DWYQEJEF4FHSG6QMAP5PU4KZ","target":"record","payload":{"canonical_record":{"source":{"id":"2607.01859","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-02T08:17:57Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"2b7c5c6a10e6d305504f33c41f26de4235e5f638c55e96c6ac2e2b1d6cd4a6bf","abstract_canon_sha256":"a450899c7d70ddca13647262070e5c90a1989b95c9a9a281b3deee90def965a3"},"schema_version":"1.0"},"canonical_sha256":"af076c4089217853c8de8300febe9c565079391cab672b0a12314d2b5213ed3d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-03T01:17:31.815533Z","signature_b64":"52R+gzQvRk7WwRWsIMLV4c2CkPB29fnkoMEyprVzpZUlXn/yG1yN7Idiv+/WlSbPHHBJCp8tTRTwIOTgkovjCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"af076c4089217853c8de8300febe9c565079391cab672b0a12314d2b5213ed3d","last_reissued_at":"2026-07-03T01:17:31.815175Z","signature_status":"signed_v1","first_computed_at":"2026-07-03T01:17:31.815175Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2607.01859","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":"2026-07-03T01:17:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PlV7S4i+IMri37z5eWHbl1Rc4narhpvXI+uFl6X1IdjPtntSRMxDWqTj1m5rAU5m4vkNsJxJzVHQyOpRFVuuCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T18:32:18.577031Z"},"content_sha256":"eb257f8271f9ee994c6cad51d70fa423f3049d6cfad4af754ef9840d597e7ad8","schema_version":"1.0","event_id":"sha256:eb257f8271f9ee994c6cad51d70fa423f3049d6cfad4af754ef9840d597e7ad8"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:V4DWYQEJEF4FHSG6QMAP5PU4KZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Safety Targeted Embedding Exploit via Refinement","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Joshua Adrian Cahyono","submitted_at":"2026-07-02T08:17:57Z","abstract_excerpt":"Safety training for large language models (LLMs) is conducted predominantly in English, leaving uncertain how well safety mechanisms generalize to low-resource languages and mixed-language code-switching. We show that this creates an epistemic gap in which models confidently generate harmful responses for inputs that fall outside the distribution of their safety training. To study this phenomenon, we introduce STEER (Safety Targeted Embedding Exploit via Refinement), a gradient-guided attack that identifies words contributing most strongly to the model's refusal behavior and iteratively transl"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.01859","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/2607.01859/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":"2026-07-03T01:17:31Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"AQFjXJJSo+GjhNjJPL6pADoNg8FtzYKNLLBTYSnRDQqPF0UxoT1gMLPHg37YJhT09FCIgKemVOyDQxck+ZibDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-03T18:32:18.577431Z"},"content_sha256":"f4f7a49e4133d2ab5d80e7e7bb1e88d8315e991f5a30d49e4a056110c336aa2d","schema_version":"1.0","event_id":"sha256:f4f7a49e4133d2ab5d80e7e7bb1e88d8315e991f5a30d49e4a056110c336aa2d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/bundle.json","state_url":"https://pith.science/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/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":"2026-07-03T18:32:18Z","links":{"resolver":"https://pith.science/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ","bundle":"https://pith.science/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/bundle.json","state":"https://pith.science/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/V4DWYQEJEF4FHSG6QMAP5PU4KZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:V4DWYQEJEF4FHSG6QMAP5PU4KZ","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":"a450899c7d70ddca13647262070e5c90a1989b95c9a9a281b3deee90def965a3","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-02T08:17:57Z","title_canon_sha256":"2b7c5c6a10e6d305504f33c41f26de4235e5f638c55e96c6ac2e2b1d6cd4a6bf"},"schema_version":"1.0","source":{"id":"2607.01859","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2607.01859","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"arxiv_version","alias_value":"2607.01859v1","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.01859","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_12","alias_value":"V4DWYQEJEF4F","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_16","alias_value":"V4DWYQEJEF4FHSG6","created_at":"2026-07-03T01:17:31Z"},{"alias_kind":"pith_short_8","alias_value":"V4DWYQEJ","created_at":"2026-07-03T01:17:31Z"}],"graph_snapshots":[{"event_id":"sha256:f4f7a49e4133d2ab5d80e7e7bb1e88d8315e991f5a30d49e4a056110c336aa2d","target":"graph","created_at":"2026-07-03T01:17:31Z","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/2607.01859/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Safety training for large language models (LLMs) is conducted predominantly in English, leaving uncertain how well safety mechanisms generalize to low-resource languages and mixed-language code-switching. We show that this creates an epistemic gap in which models confidently generate harmful responses for inputs that fall outside the distribution of their safety training. To study this phenomenon, we introduce STEER (Safety Targeted Embedding Exploit via Refinement), a gradient-guided attack that identifies words contributing most strongly to the model's refusal behavior and iteratively transl","authors_text":"Joshua Adrian Cahyono","cross_cats":["cs.CL"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-02T08:17:57Z","title":"Safety Targeted Embedding Exploit via Refinement"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.01859","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:eb257f8271f9ee994c6cad51d70fa423f3049d6cfad4af754ef9840d597e7ad8","target":"record","created_at":"2026-07-03T01:17:31Z","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":"a450899c7d70ddca13647262070e5c90a1989b95c9a9a281b3deee90def965a3","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-02T08:17:57Z","title_canon_sha256":"2b7c5c6a10e6d305504f33c41f26de4235e5f638c55e96c6ac2e2b1d6cd4a6bf"},"schema_version":"1.0","source":{"id":"2607.01859","kind":"arxiv","version":1}},"canonical_sha256":"af076c4089217853c8de8300febe9c565079391cab672b0a12314d2b5213ed3d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"af076c4089217853c8de8300febe9c565079391cab672b0a12314d2b5213ed3d","first_computed_at":"2026-07-03T01:17:31.815175Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-03T01:17:31.815175Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"52R+gzQvRk7WwRWsIMLV4c2CkPB29fnkoMEyprVzpZUlXn/yG1yN7Idiv+/WlSbPHHBJCp8tTRTwIOTgkovjCA==","signature_status":"signed_v1","signed_at":"2026-07-03T01:17:31.815533Z","signed_message":"canonical_sha256_bytes"},"source_id":"2607.01859","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:eb257f8271f9ee994c6cad51d70fa423f3049d6cfad4af754ef9840d597e7ad8","sha256:f4f7a49e4133d2ab5d80e7e7bb1e88d8315e991f5a30d49e4a056110c336aa2d"],"state_sha256":"1493ba99c179be024ce5fa1aa5e464070be3c28accf5bde4a503e462b60ceb45"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ytIdU3nMMdndjey3erQFec2ryveATiyxRn6fJ1y2qm3RqX3JcVawfpvSB9ahqhjj2BHAsYvTjitclY92swg+CQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-03T18:32:18.579379Z","bundle_sha256":"cdd7b7e9bb3e55ee17c681a89a8d5c91dc222e5025224d79b45dbcc5f190dbf6"}}