{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:UP2A6MXFA4JD2Z2JVF6UACCIMN","short_pith_number":"pith:UP2A6MXF","schema_version":"1.0","canonical_sha256":"a3f40f32e507123d6749a97d4008486372b5cf3fd507aa1f5da31c599f8b2953","source":{"kind":"arxiv","id":"2605.28591","version":1},"attestation_state":"computed","paper":{"title":"Models That Know How Evaluations Are Designed Score Safer","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Haritz Puerto, Jonas Geiping, Katharina Deckenbach, Sahar Abdelnabi","submitted_at":"2026-05-27T15:11:35Z","abstract_excerpt":"The validity of AI safety evaluations depends on models behaving consistently across controlled and deployment settings. Prior work has identified test-time contextual cues, such as hypothetical scenarios, as a source of verbalized evaluation awareness and subsequent behavioral shift. In this paper, we investigate a potential explanation of this phenomenon: evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations. Similar to dataset contamination, where benchmark exposure leads to higher performance through memorization, we hypothesiz"},"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":"2605.28591","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-27T15:11:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cb53a1efe8fa3a1897bf15c9d0cbe4e2402b801c09077cfdcaf2391c5a9f1fbb","abstract_canon_sha256":"d32535f5454c65c831bd0dfb28a41b24992ff8dac576611b2803aaa86beff963"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T02:04:57.264938Z","signature_b64":"XEGshiucUAfgljdTicLHuy/4kwKrnTcZcEq+C4HlxqzB2XN/RFjFasI5hqIZ0c0QemmQd4UUNj2+2TfVPu55BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3f40f32e507123d6749a97d4008486372b5cf3fd507aa1f5da31c599f8b2953","last_reissued_at":"2026-05-28T02:04:57.264511Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T02:04:57.264511Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Models That Know How Evaluations Are Designed Score Safer","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Haritz Puerto, Jonas Geiping, Katharina Deckenbach, Sahar Abdelnabi","submitted_at":"2026-05-27T15:11:35Z","abstract_excerpt":"The validity of AI safety evaluations depends on models behaving consistently across controlled and deployment settings. Prior work has identified test-time contextual cues, such as hypothetical scenarios, as a source of verbalized evaluation awareness and subsequent behavioral shift. In this paper, we investigate a potential explanation of this phenomenon: evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations. Similar to dataset contamination, where benchmark exposure leads to higher performance through memorization, we hypothesiz"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28591","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/2605.28591/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":"2605.28591","created_at":"2026-05-28T02:04:57.264574+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28591v1","created_at":"2026-05-28T02:04:57.264574+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28591","created_at":"2026-05-28T02:04:57.264574+00:00"},{"alias_kind":"pith_short_12","alias_value":"UP2A6MXFA4JD","created_at":"2026-05-28T02:04:57.264574+00:00"},{"alias_kind":"pith_short_16","alias_value":"UP2A6MXFA4JD2Z2J","created_at":"2026-05-28T02:04:57.264574+00:00"},{"alias_kind":"pith_short_8","alias_value":"UP2A6MXF","created_at":"2026-05-28T02:04:57.264574+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/UP2A6MXFA4JD2Z2JVF6UACCIMN","json":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN.json","graph_json":"https://pith.science/api/pith-number/UP2A6MXFA4JD2Z2JVF6UACCIMN/graph.json","events_json":"https://pith.science/api/pith-number/UP2A6MXFA4JD2Z2JVF6UACCIMN/events.json","paper":"https://pith.science/paper/UP2A6MXF"},"agent_actions":{"view_html":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN","download_json":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN.json","view_paper":"https://pith.science/paper/UP2A6MXF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28591&json=true","fetch_graph":"https://pith.science/api/pith-number/UP2A6MXFA4JD2Z2JVF6UACCIMN/graph.json","fetch_events":"https://pith.science/api/pith-number/UP2A6MXFA4JD2Z2JVF6UACCIMN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN/action/storage_attestation","attest_author":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN/action/author_attestation","sign_citation":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN/action/citation_signature","submit_replication":"https://pith.science/pith/UP2A6MXFA4JD2Z2JVF6UACCIMN/action/replication_record"}},"created_at":"2026-05-28T02:04:57.264574+00:00","updated_at":"2026-05-28T02:04:57.264574+00:00"}