{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GFIFYHZZQLCX67PTXEO7TYD35G","short_pith_number":"pith:GFIFYHZZ","schema_version":"1.0","canonical_sha256":"31505c1f3982c57f7df3b91df9e07be9b693fb3f94480e7e556651e8b2d6b847","source":{"kind":"arxiv","id":"2602.01425","version":2},"attestation_state":"computed","paper":{"title":"One Probe Won't Catch Them All: Towards Targeted Deception Detection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Devina Jain, Joseph Bloom, Satvik Golechha, Shivam Arora, Vikram Natarajan","submitted_at":"2026-02-01T20:18:11Z","abstract_excerpt":"Linear probes are a promising approach for monitoring AI systems for deceptive behaviour. Previous work has shown that a linear classifier trained on a contrastive instruction pair and a simple dataset can achieve good performance. However, these probes exhibit notable failures even in straightforward scenarios, including spurious correlations and false positives on non-deceptive responses. In this paper, we demonstrate that deception detection is inherently heterogeneous: while a single universal probe achieves modest improvements (+0.032 AUC), post-hoc oracle analysis reveals substantially h"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2602.01425","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-02-01T20:18:11Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"6d3b87413fa46fdfaa842ffcf73d08a38d79ff4247ab10e464beac1757024557","abstract_canon_sha256":"031e02220c1b3b424ad7affcbe95747d2aa9e19bf8c6ce964009ae506a68ed95"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:18.258882Z","signature_b64":"rznwDMPHzZF7/EawXndWTXhn/5iirdL5WqR5e0jHL1DNf8vcmDNzU/sTIIWpirxGckm3yvO3MvfNP4l9LWPOAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"31505c1f3982c57f7df3b91df9e07be9b693fb3f94480e7e556651e8b2d6b847","last_reissued_at":"2026-06-19T16:12:18.258466Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:18.258466Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"One Probe Won't Catch Them All: Towards Targeted Deception Detection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Devina Jain, Joseph Bloom, Satvik Golechha, Shivam Arora, Vikram Natarajan","submitted_at":"2026-02-01T20:18:11Z","abstract_excerpt":"Linear probes are a promising approach for monitoring AI systems for deceptive behaviour. Previous work has shown that a linear classifier trained on a contrastive instruction pair and a simple dataset can achieve good performance. However, these probes exhibit notable failures even in straightforward scenarios, including spurious correlations and false positives on non-deceptive responses. In this paper, we demonstrate that deception detection is inherently heterogeneous: while a single universal probe achieves modest improvements (+0.032 AUC), post-hoc oracle analysis reveals substantially h"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.01425","kind":"arxiv","version":2},"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/2602.01425/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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2602.01425","created_at":"2026-06-19T16:12:18.258521+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.01425v2","created_at":"2026-06-19T16:12:18.258521+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.01425","created_at":"2026-06-19T16:12:18.258521+00:00"},{"alias_kind":"pith_short_12","alias_value":"GFIFYHZZQLCX","created_at":"2026-06-19T16:12:18.258521+00:00"},{"alias_kind":"pith_short_16","alias_value":"GFIFYHZZQLCX67PT","created_at":"2026-06-19T16:12:18.258521+00:00"},{"alias_kind":"pith_short_8","alias_value":"GFIFYHZZ","created_at":"2026-06-19T16:12:18.258521+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2606.12618","citing_title":"\"Did you lie?\" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms","ref_index":84,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09391","citing_title":"Do Linear Probes Generalize Better in Persona Coordinates?","ref_index":62,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09391","citing_title":"Do Linear Probes Generalize Better in Persona Coordinates?","ref_index":63,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G","json":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G.json","graph_json":"https://pith.science/api/pith-number/GFIFYHZZQLCX67PTXEO7TYD35G/graph.json","events_json":"https://pith.science/api/pith-number/GFIFYHZZQLCX67PTXEO7TYD35G/events.json","paper":"https://pith.science/paper/GFIFYHZZ"},"agent_actions":{"view_html":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G","download_json":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G.json","view_paper":"https://pith.science/paper/GFIFYHZZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.01425&json=true","fetch_graph":"https://pith.science/api/pith-number/GFIFYHZZQLCX67PTXEO7TYD35G/graph.json","fetch_events":"https://pith.science/api/pith-number/GFIFYHZZQLCX67PTXEO7TYD35G/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G/action/storage_attestation","attest_author":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G/action/author_attestation","sign_citation":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G/action/citation_signature","submit_replication":"https://pith.science/pith/GFIFYHZZQLCX67PTXEO7TYD35G/action/replication_record"}},"created_at":"2026-06-19T16:12:18.258521+00:00","updated_at":"2026-06-19T16:12:18.258521+00:00"}