{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:26VRZ3IWT6PILFXFYZO3NHXXW3","short_pith_number":"pith:26VRZ3IW","schema_version":"1.0","canonical_sha256":"d7ab1ced169f9e8596e5c65db69ef7b6d4337513a760d1ec3974a9e0f21e8665","source":{"kind":"arxiv","id":"2607.23379","version":1},"attestation_state":"computed","paper":{"title":"When Activation Oracles Learn Not to Read: Concept-Specific Blind Spots in Fine-Tuned Oracles","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Tatiana Gaintseva, Tobias Bersia","submitted_at":"2026-07-25T21:58:17Z","abstract_excerpt":"Activation Oracles (AOs) are language models trained to answer natural-language questions about another model's internal activations. They offer a flexible interface for reading hidden information from model states, especially when relevant information is internally represented but absent or incomplete in visible behavior. However, AOs are themselves learned systems: their answers are shaped by training data, objectives, and learned reporting behavior, rather than being neutral readouts of represented information. We study this in a controlled Taboo Word Guessing setting, where subject models "},"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":"2607.23379","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-07-25T21:58:17Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"245a101bc9b4f43c55c905ff5931778fa0b95d7fdb68d6b12e89e9beb83d22c4","abstract_canon_sha256":"9c375446640f748c0c20fb94cd51e006ffe05c3fa304dde964db6a8eef424722"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T01:22:49.887868Z","signature_b64":"UUR0h9yrM8lZYE+xK+eAls20s9xeQ0waCvdReQB3nQk3TYpcMCVpAZXT3x8/IFUJIWaIDcxpcfeiJAms+tSDDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d7ab1ced169f9e8596e5c65db69ef7b6d4337513a760d1ec3974a9e0f21e8665","last_reissued_at":"2026-07-28T01:22:49.887028Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T01:22:49.887028Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Activation Oracles Learn Not to Read: Concept-Specific Blind Spots in Fine-Tuned Oracles","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Tatiana Gaintseva, Tobias Bersia","submitted_at":"2026-07-25T21:58:17Z","abstract_excerpt":"Activation Oracles (AOs) are language models trained to answer natural-language questions about another model's internal activations. They offer a flexible interface for reading hidden information from model states, especially when relevant information is internally represented but absent or incomplete in visible behavior. However, AOs are themselves learned systems: their answers are shaped by training data, objectives, and learned reporting behavior, rather than being neutral readouts of represented information. We study this in a controlled Taboo Word Guessing setting, where subject models "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.23379","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/2607.23379/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":"2607.23379","created_at":"2026-07-28T01:22:49.887449+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.23379v1","created_at":"2026-07-28T01:22:49.887449+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.23379","created_at":"2026-07-28T01:22:49.887449+00:00"},{"alias_kind":"pith_short_12","alias_value":"26VRZ3IWT6PI","created_at":"2026-07-28T01:22:49.887449+00:00"},{"alias_kind":"pith_short_16","alias_value":"26VRZ3IWT6PILFXF","created_at":"2026-07-28T01:22:49.887449+00:00"},{"alias_kind":"pith_short_8","alias_value":"26VRZ3IW","created_at":"2026-07-28T01:22:49.887449+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3","json":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3.json","graph_json":"https://pith.science/api/pith-number/26VRZ3IWT6PILFXFYZO3NHXXW3/graph.json","events_json":"https://pith.science/api/pith-number/26VRZ3IWT6PILFXFYZO3NHXXW3/events.json","paper":"https://pith.science/paper/26VRZ3IW"},"agent_actions":{"view_html":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3","download_json":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3.json","view_paper":"https://pith.science/paper/26VRZ3IW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.23379&json=true","fetch_graph":"https://pith.science/api/pith-number/26VRZ3IWT6PILFXFYZO3NHXXW3/graph.json","fetch_events":"https://pith.science/api/pith-number/26VRZ3IWT6PILFXFYZO3NHXXW3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3/action/storage_attestation","attest_author":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3/action/author_attestation","sign_citation":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3/action/citation_signature","submit_replication":"https://pith.science/pith/26VRZ3IWT6PILFXFYZO3NHXXW3/action/replication_record"}},"created_at":"2026-07-28T01:22:49.887449+00:00","updated_at":"2026-07-28T01:22:49.887449+00:00"}