{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Z3TWDYNB6WJE7OBTP2DB2FXJUV","short_pith_number":"pith:Z3TWDYNB","schema_version":"1.0","canonical_sha256":"cee761e1a1f5924fb8337e861d16e9a579886fcfafe54ff4acc352c101bc3b86","source":{"kind":"arxiv","id":"2605.26574","version":1},"attestation_state":"computed","paper":{"title":"GradSentry: Gradient Spectral Entropy for Backdoor Sample Filtering in Large Language Model Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Gongshen Liu, Haodong Zhao, Tianhang Zhao, Tianyi Xu, Zhuosheng Zhang","submitted_at":"2026-05-26T05:48:01Z","abstract_excerpt":"Fine-tuning Large Language Models with untrusted data exposes models to backdoor attacks, where poisoned samples cause targeted misbehavior. Existing sample-filtering defenses rely on clustering, which requires sufficient data and can fail at extreme poison ratios. We propose GradSentry ({Grad}ient {Sentry}), a backdoor sample filtering method based on the spectral entropy of per-sample gradients. Our key finding is that poisoned samples produce gradients with higher spectral entropy compared to clean samples. GradSentry captures output-altering backdoor signatures using per-sample gradient sp"},"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":"2605.26574","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-05-26T05:48:01Z","cross_cats_sorted":[],"title_canon_sha256":"f3136ffa4207e451b79af2c6841ebffa067b247e853485d4be10e72fa20c02e7","abstract_canon_sha256":"8cb59459cf7d2ad981902a0fe18a884c712772bfc0f95d23658324357259dc45"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:05:26.742884Z","signature_b64":"8Z52RiQVsb2uZBoDJh5FompuryB2+jp4HBd+jy4bExhMmOlRplHDzK/EHS+sKba8MxgMV85ODbHIf/d5PxsHCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cee761e1a1f5924fb8337e861d16e9a579886fcfafe54ff4acc352c101bc3b86","last_reissued_at":"2026-05-27T01:05:26.742410Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:05:26.742410Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GradSentry: Gradient Spectral Entropy for Backdoor Sample Filtering in Large Language Model Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Gongshen Liu, Haodong Zhao, Tianhang Zhao, Tianyi Xu, Zhuosheng Zhang","submitted_at":"2026-05-26T05:48:01Z","abstract_excerpt":"Fine-tuning Large Language Models with untrusted data exposes models to backdoor attacks, where poisoned samples cause targeted misbehavior. Existing sample-filtering defenses rely on clustering, which requires sufficient data and can fail at extreme poison ratios. We propose GradSentry ({Grad}ient {Sentry}), a backdoor sample filtering method based on the spectral entropy of per-sample gradients. Our key finding is that poisoned samples produce gradients with higher spectral entropy compared to clean samples. GradSentry captures output-altering backdoor signatures using per-sample gradient sp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26574","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/2605.26574/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":"2605.26574","created_at":"2026-05-27T01:05:26.742478+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26574v1","created_at":"2026-05-27T01:05:26.742478+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26574","created_at":"2026-05-27T01:05:26.742478+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z3TWDYNB6WJE","created_at":"2026-05-27T01:05:26.742478+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z3TWDYNB6WJE7OBT","created_at":"2026-05-27T01:05:26.742478+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z3TWDYNB","created_at":"2026-05-27T01:05:26.742478+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/Z3TWDYNB6WJE7OBTP2DB2FXJUV","json":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV.json","graph_json":"https://pith.science/api/pith-number/Z3TWDYNB6WJE7OBTP2DB2FXJUV/graph.json","events_json":"https://pith.science/api/pith-number/Z3TWDYNB6WJE7OBTP2DB2FXJUV/events.json","paper":"https://pith.science/paper/Z3TWDYNB"},"agent_actions":{"view_html":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV","download_json":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV.json","view_paper":"https://pith.science/paper/Z3TWDYNB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26574&json=true","fetch_graph":"https://pith.science/api/pith-number/Z3TWDYNB6WJE7OBTP2DB2FXJUV/graph.json","fetch_events":"https://pith.science/api/pith-number/Z3TWDYNB6WJE7OBTP2DB2FXJUV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV/action/storage_attestation","attest_author":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV/action/author_attestation","sign_citation":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV/action/citation_signature","submit_replication":"https://pith.science/pith/Z3TWDYNB6WJE7OBTP2DB2FXJUV/action/replication_record"}},"created_at":"2026-05-27T01:05:26.742478+00:00","updated_at":"2026-05-27T01:05:26.742478+00:00"}