{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2SI2JNX2UVQL2WWS6EGFKKHBOD","short_pith_number":"pith:2SI2JNX2","schema_version":"1.0","canonical_sha256":"d491a4b6faa560bd5ad2f10c5528e170d03d2b1de7d9eb87997e7cb87a752102","source":{"kind":"arxiv","id":"2505.12368","version":2},"attestation_state":"computed","paper":{"title":"CAPTURE: Context-Aware Prompt Injection Testing and Robustness Enhancement","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gauri Kholkar, Ratinder Ahuja","submitted_at":"2025-05-18T11:14:14Z","abstract_excerpt":"Prompt injection remains a major security risk for large language models. However, the efficacy of existing guardrail models in context-aware settings remains underexplored, as they often rely on static attack benchmarks. Additionally, they have over-defense tendencies. We introduce CAPTURE, a novel context-aware benchmark assessing both attack detection and over-defense tendencies with minimal in-domain examples. Our experiments reveal that current prompt injection guardrail models suffer from high false negatives in adversarial cases and excessive false positives in benign scenarios, highlig"},"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":"2505.12368","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-18T11:14:14Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5caeed117a3c8921c9b703eff1d9189a926b1ceaea2f38f73db2ebdc5ac6965c","abstract_canon_sha256":"a9e5266e04060bedc01a263d8a03da78345a0ae13282c2682262290b4320662d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:22:40.277862Z","signature_b64":"cq5k5aLsvgDVgpMm72KrtzX2MYrz7oF3A9/44XsyzJ6ImLQf0X96cLDhKjbUCIsiWMfODsrNpy0LV8V6JTUkAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d491a4b6faa560bd5ad2f10c5528e170d03d2b1de7d9eb87997e7cb87a752102","last_reissued_at":"2026-07-05T11:22:40.277371Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:22:40.277371Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CAPTURE: Context-Aware Prompt Injection Testing and Robustness Enhancement","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gauri Kholkar, Ratinder Ahuja","submitted_at":"2025-05-18T11:14:14Z","abstract_excerpt":"Prompt injection remains a major security risk for large language models. However, the efficacy of existing guardrail models in context-aware settings remains underexplored, as they often rely on static attack benchmarks. Additionally, they have over-defense tendencies. We introduce CAPTURE, a novel context-aware benchmark assessing both attack detection and over-defense tendencies with minimal in-domain examples. Our experiments reveal that current prompt injection guardrail models suffer from high false negatives in adversarial cases and excessive false positives in benign scenarios, highlig"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12368","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/2505.12368/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":"2505.12368","created_at":"2026-07-05T11:22:40.277429+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.12368v2","created_at":"2026-07-05T11:22:40.277429+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12368","created_at":"2026-07-05T11:22:40.277429+00:00"},{"alias_kind":"pith_short_12","alias_value":"2SI2JNX2UVQL","created_at":"2026-07-05T11:22:40.277429+00:00"},{"alias_kind":"pith_short_16","alias_value":"2SI2JNX2UVQL2WWS","created_at":"2026-07-05T11:22:40.277429+00:00"},{"alias_kind":"pith_short_8","alias_value":"2SI2JNX2","created_at":"2026-07-05T11:22:40.277429+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/2SI2JNX2UVQL2WWS6EGFKKHBOD","json":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD.json","graph_json":"https://pith.science/api/pith-number/2SI2JNX2UVQL2WWS6EGFKKHBOD/graph.json","events_json":"https://pith.science/api/pith-number/2SI2JNX2UVQL2WWS6EGFKKHBOD/events.json","paper":"https://pith.science/paper/2SI2JNX2"},"agent_actions":{"view_html":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD","download_json":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD.json","view_paper":"https://pith.science/paper/2SI2JNX2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.12368&json=true","fetch_graph":"https://pith.science/api/pith-number/2SI2JNX2UVQL2WWS6EGFKKHBOD/graph.json","fetch_events":"https://pith.science/api/pith-number/2SI2JNX2UVQL2WWS6EGFKKHBOD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD/action/storage_attestation","attest_author":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD/action/author_attestation","sign_citation":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD/action/citation_signature","submit_replication":"https://pith.science/pith/2SI2JNX2UVQL2WWS6EGFKKHBOD/action/replication_record"}},"created_at":"2026-07-05T11:22:40.277429+00:00","updated_at":"2026-07-05T11:22:40.277429+00:00"}