{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:4DSNBLEYQTBMN3YT433QXWOEJ2","short_pith_number":"pith:4DSNBLEY","canonical_record":{"source":{"id":"2504.07831","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-10T15:07:10Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"c12640f3b30d52e3c220d4f148e258b4cccbe755d747bad92c8f19b0f3d98318","abstract_canon_sha256":"24e5980dedc505bad599aee97238c7dc282e68819c38b6256ce0a06c31749ea3"},"schema_version":"1.0"},"canonical_sha256":"e0e4d0ac9884c2c6ef13e6f70bd9c44e8c4a623ef657bc40ea61226ee8bc540e","source":{"kind":"arxiv","id":"2504.07831","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.07831","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"arxiv_version","alias_value":"2504.07831v1","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07831","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_12","alias_value":"4DSNBLEYQTBM","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_16","alias_value":"4DSNBLEYQTBMN3YT","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_8","alias_value":"4DSNBLEY","created_at":"2026-07-05T10:47:15Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:4DSNBLEYQTBMN3YT433QXWOEJ2","target":"record","payload":{"canonical_record":{"source":{"id":"2504.07831","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-10T15:07:10Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"c12640f3b30d52e3c220d4f148e258b4cccbe755d747bad92c8f19b0f3d98318","abstract_canon_sha256":"24e5980dedc505bad599aee97238c7dc282e68819c38b6256ce0a06c31749ea3"},"schema_version":"1.0"},"canonical_sha256":"e0e4d0ac9884c2c6ef13e6f70bd9c44e8c4a623ef657bc40ea61226ee8bc540e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:47:15.561393Z","signature_b64":"Kp4l2sQ/Fft8t9t+N0i0dyLdgRrz2r/7zuwdOOAjjRQJrx7Iz6AO/jouS199pdbgte3sOIOYwykhIOE2jhwmDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e0e4d0ac9884c2c6ef13e6f70bd9c44e8c4a623ef657bc40ea61226ee8bc540e","last_reissued_at":"2026-07-05T10:47:15.560917Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:47:15.560917Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2504.07831","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":"2026-07-05T10:47:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"U/4SHoQ+3HSq21ZwCjXVoiY2YgxradtG3L6HcAxAiSf0Fnyxzonafwe7MMdrcdHsFqBadm3WzjMo5rkg4XsPBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-19T09:16:01.884364Z"},"content_sha256":"7a8b01eaead8e95aed69114f438823f44adf48db8d7c7f1f651b32bd8472fe92","schema_version":"1.0","event_id":"sha256:7a8b01eaead8e95aed69114f438823f44adf48db8d7c7f1f651b32bd8472fe92"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:4DSNBLEYQTBMN3YT433QXWOEJ2","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight Systems","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Mateusz Dziemian, Natalia P\\'erez-Campanero Antol\\'in, Simon Lermen","submitted_at":"2025-04-10T15:07:10Z","abstract_excerpt":"We demonstrate how AI agents can coordinate to deceive oversight systems using automated interpretability of neural networks. Using sparse autoencoders (SAEs) as our experimental framework, we show that language models (Llama, DeepSeek R1, and Claude 3.7 Sonnet) can generate deceptive explanations that evade detection. Our agents employ steganographic methods to hide information in seemingly innocent explanations, successfully fooling oversight models while achieving explanation quality comparable to reference labels. We further find that models can scheme to develop deceptive strategies when "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07831","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/2504.07831/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":"2026-07-05T10:47:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"KgH3qmyBIERgSOBJC17Z08Fbyp/oUt/tNO0RN5bixxsCdG0zLUgoCbFqWSTGYfnz6KF07zcClfMETYky8kYyCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-19T09:16:01.884899Z"},"content_sha256":"ca9450b60070788afc67094fae276ec356216061bf3a52cef2e94308aa39ed1b","schema_version":"1.0","event_id":"sha256:ca9450b60070788afc67094fae276ec356216061bf3a52cef2e94308aa39ed1b"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/bundle.json","state_url":"https://pith.science/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/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":"2026-08-19T09:16:01Z","links":{"resolver":"https://pith.science/pith/4DSNBLEYQTBMN3YT433QXWOEJ2","bundle":"https://pith.science/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/bundle.json","state":"https://pith.science/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/state.json","well_known_bundle":"https://pith.science/.well-known/pith/4DSNBLEYQTBMN3YT433QXWOEJ2/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:4DSNBLEYQTBMN3YT433QXWOEJ2","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":"24e5980dedc505bad599aee97238c7dc282e68819c38b6256ce0a06c31749ea3","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-10T15:07:10Z","title_canon_sha256":"c12640f3b30d52e3c220d4f148e258b4cccbe755d747bad92c8f19b0f3d98318"},"schema_version":"1.0","source":{"id":"2504.07831","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.07831","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"arxiv_version","alias_value":"2504.07831v1","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07831","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_12","alias_value":"4DSNBLEYQTBM","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_16","alias_value":"4DSNBLEYQTBMN3YT","created_at":"2026-07-05T10:47:15Z"},{"alias_kind":"pith_short_8","alias_value":"4DSNBLEY","created_at":"2026-07-05T10:47:15Z"}],"graph_snapshots":[{"event_id":"sha256:ca9450b60070788afc67094fae276ec356216061bf3a52cef2e94308aa39ed1b","target":"graph","created_at":"2026-07-05T10:47:15Z","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/2504.07831/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We demonstrate how AI agents can coordinate to deceive oversight systems using automated interpretability of neural networks. Using sparse autoencoders (SAEs) as our experimental framework, we show that language models (Llama, DeepSeek R1, and Claude 3.7 Sonnet) can generate deceptive explanations that evade detection. Our agents employ steganographic methods to hide information in seemingly innocent explanations, successfully fooling oversight models while achieving explanation quality comparable to reference labels. We further find that models can scheme to develop deceptive strategies when ","authors_text":"Mateusz Dziemian, Natalia P\\'erez-Campanero Antol\\'in, Simon Lermen","cross_cats":["cs.CL"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-10T15:07:10Z","title":"Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight Systems"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07831","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":"sha256:7a8b01eaead8e95aed69114f438823f44adf48db8d7c7f1f651b32bd8472fe92","target":"record","created_at":"2026-07-05T10:47:15Z","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":"24e5980dedc505bad599aee97238c7dc282e68819c38b6256ce0a06c31749ea3","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-10T15:07:10Z","title_canon_sha256":"c12640f3b30d52e3c220d4f148e258b4cccbe755d747bad92c8f19b0f3d98318"},"schema_version":"1.0","source":{"id":"2504.07831","kind":"arxiv","version":1}},"canonical_sha256":"e0e4d0ac9884c2c6ef13e6f70bd9c44e8c4a623ef657bc40ea61226ee8bc540e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e0e4d0ac9884c2c6ef13e6f70bd9c44e8c4a623ef657bc40ea61226ee8bc540e","first_computed_at":"2026-07-05T10:47:15.560917Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:47:15.560917Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Kp4l2sQ/Fft8t9t+N0i0dyLdgRrz2r/7zuwdOOAjjRQJrx7Iz6AO/jouS199pdbgte3sOIOYwykhIOE2jhwmDg==","signature_status":"signed_v1","signed_at":"2026-07-05T10:47:15.561393Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.07831","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:7a8b01eaead8e95aed69114f438823f44adf48db8d7c7f1f651b32bd8472fe92","sha256:ca9450b60070788afc67094fae276ec356216061bf3a52cef2e94308aa39ed1b"],"state_sha256":"9d2f3948f0c4e6cec775cfed2c4bfe8551aea461c6045f4663500baf6a55e68f"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"FEtZKZu6VnigSGVvDwWSrIftCERvqRpQP4ZLavn77f3NrcVB/rgsjngEla0yaG7ML8iR71Eo4sxBBmty8ulcAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-19T09:16:01.889590Z","bundle_sha256":"f59b081ff54e411914fbd4ec238526fde076b83adf38bef6b661727f899d5438"}}