{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:A5V66ZEG5HOUKSQKTWBYBW4FW4","short_pith_number":"pith:A5V66ZEG","schema_version":"1.0","canonical_sha256":"076bef6486e9dd454a0a9d8380db85b726b5705b10ae2e2678a8a9cafba4156f","source":{"kind":"arxiv","id":"2506.04390","version":2},"attestation_state":"computed","paper":{"title":"Through the Stealth Lens: Attention-Aware Defenses Against Poisoning in RAG","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Ashish Hooda, Krishnamurthy Dj Dvijotham, Nils Palumbo, Sarthak Choudhary, Somesh Jha","submitted_at":"2025-06-04T19:15:09Z","abstract_excerpt":"Retrieval-augmented generation (RAG) systems are vulnerable to attacks that inject poisoned passages into the retrieved context, even at low corruption rates. We show that existing attacks are not designed to be stealthy, allowing reliable detection and mitigation. We formalize a distinguishability-based security game to quantify stealth for such attacks. If a few poisoned passages control the response, they must bias the inference process more than the benign ones, inherently compromising stealth. This motivates analyzing intermediate signals of LLMs, such as attention weights, to approximate"},"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":"2506.04390","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-06-04T19:15:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"85a159c099ac9067c1ef87584039d6e654ae79f5851a3e1104fe76a6386499a3","abstract_canon_sha256":"af4cd6656325970d6df908c1fbf0d37cbb4eda8aa828adf23288c8f234342c9d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:04.712804Z","signature_b64":"FIOvYZzJwtyl8nz9mbOLubIiWuHzJefjYkJxfadvR5DXDo+CAQZXuitsONCImOpRl7CWA8Az1mDI3RMIg77vCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"076bef6486e9dd454a0a9d8380db85b726b5705b10ae2e2678a8a9cafba4156f","last_reissued_at":"2026-05-25T02:01:04.711951Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:04.711951Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Through the Stealth Lens: Attention-Aware Defenses Against Poisoning in RAG","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Ashish Hooda, Krishnamurthy Dj Dvijotham, Nils Palumbo, Sarthak Choudhary, Somesh Jha","submitted_at":"2025-06-04T19:15:09Z","abstract_excerpt":"Retrieval-augmented generation (RAG) systems are vulnerable to attacks that inject poisoned passages into the retrieved context, even at low corruption rates. We show that existing attacks are not designed to be stealthy, allowing reliable detection and mitigation. We formalize a distinguishability-based security game to quantify stealth for such attacks. If a few poisoned passages control the response, they must bias the inference process more than the benign ones, inherently compromising stealth. This motivates analyzing intermediate signals of LLMs, such as attention weights, to approximate"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04390","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/2506.04390/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":"2506.04390","created_at":"2026-05-25T02:01:04.712091+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.04390v2","created_at":"2026-05-25T02:01:04.712091+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04390","created_at":"2026-05-25T02:01:04.712091+00:00"},{"alias_kind":"pith_short_12","alias_value":"A5V66ZEG5HOU","created_at":"2026-05-25T02:01:04.712091+00:00"},{"alias_kind":"pith_short_16","alias_value":"A5V66ZEG5HOUKSQK","created_at":"2026-05-25T02:01:04.712091+00:00"},{"alias_kind":"pith_short_8","alias_value":"A5V66ZEG","created_at":"2026-05-25T02:01:04.712091+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.01782","citing_title":"Needle-in-RAG: Prompt-Conditioned Character-Level Traceback of Poisoned Spans in Retrieved Evidence","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2604.20932","citing_title":"Adaptive Defense Orchestration for RAG: A Sentinel-Strategist Architecture against Multi-Vector Attacks","ref_index":17,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4","json":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4.json","graph_json":"https://pith.science/api/pith-number/A5V66ZEG5HOUKSQKTWBYBW4FW4/graph.json","events_json":"https://pith.science/api/pith-number/A5V66ZEG5HOUKSQKTWBYBW4FW4/events.json","paper":"https://pith.science/paper/A5V66ZEG"},"agent_actions":{"view_html":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4","download_json":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4.json","view_paper":"https://pith.science/paper/A5V66ZEG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.04390&json=true","fetch_graph":"https://pith.science/api/pith-number/A5V66ZEG5HOUKSQKTWBYBW4FW4/graph.json","fetch_events":"https://pith.science/api/pith-number/A5V66ZEG5HOUKSQKTWBYBW4FW4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4/action/storage_attestation","attest_author":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4/action/author_attestation","sign_citation":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4/action/citation_signature","submit_replication":"https://pith.science/pith/A5V66ZEG5HOUKSQKTWBYBW4FW4/action/replication_record"}},"created_at":"2026-05-25T02:01:04.712091+00:00","updated_at":"2026-05-25T02:01:04.712091+00:00"}