{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:FAMEVBUUXGCEUWSGGFC4FRB5BS","short_pith_number":"pith:FAMEVBUU","schema_version":"1.0","canonical_sha256":"28184a8694b9844a5a463145c2c43d0c8f532c73b8e2694040bd686236658dcd","source":{"kind":"arxiv","id":"2607.15081","version":1},"attestation_state":"computed","paper":{"title":"DataShield: Uncovering Risky Fine-Tuning Data Across LLMs Through Consensus Subspace Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Chaochao Lu, Di Hong, Jielong Chen, Kui Ren, Liang He, Tianhang Zheng, Weiwei Qi, Zefeng Wu, Zhan Qin","submitted_at":"2026-07-16T14:51:42Z","abstract_excerpt":"Fine-tuning large language models (LLMs) on domain-specific datasets has become a standard paradigm for adapting LLMs to specialized applications. However, recent work has shown that even fine-tuning on benign task-specific data can substantially weaken the safety capabilities of LLMs. While existing efforts have made progress in identifying data responsible for safety degradation, they usually rely on a single mean vector computed over a specific model with its tokenizer to represent the safety direction, which limits both the effectiveness and transferability of their risk assessment measure"},"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":"2607.15081","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-07-16T14:51:42Z","cross_cats_sorted":[],"title_canon_sha256":"c736e7d986ab8868591af5523fecee8522e172dc9da782786577a3f34bbaf4a2","abstract_canon_sha256":"f8233250a6b06c31e9f29ab805566647813e146afc7d00f6bfaa4a0e256e21ab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T01:22:08.558889Z","signature_b64":"Mb26z5de76LTuT/kMSm8aYvfsIkiF+n1wdHcT9RSWUkmIjU87pg99I6tGtJyGzkfewbggUIxn1ttrKaYCwILCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"28184a8694b9844a5a463145c2c43d0c8f532c73b8e2694040bd686236658dcd","last_reissued_at":"2026-07-17T01:22:08.558062Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T01:22:08.558062Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DataShield: Uncovering Risky Fine-Tuning Data Across LLMs Through Consensus Subspace Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Chaochao Lu, Di Hong, Jielong Chen, Kui Ren, Liang He, Tianhang Zheng, Weiwei Qi, Zefeng Wu, Zhan Qin","submitted_at":"2026-07-16T14:51:42Z","abstract_excerpt":"Fine-tuning large language models (LLMs) on domain-specific datasets has become a standard paradigm for adapting LLMs to specialized applications. However, recent work has shown that even fine-tuning on benign task-specific data can substantially weaken the safety capabilities of LLMs. While existing efforts have made progress in identifying data responsible for safety degradation, they usually rely on a single mean vector computed over a specific model with its tokenizer to represent the safety direction, which limits both the effectiveness and transferability of their risk assessment measure"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.15081","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/2607.15081/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":"2607.15081","created_at":"2026-07-17T01:22:08.558495+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.15081v1","created_at":"2026-07-17T01:22:08.558495+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.15081","created_at":"2026-07-17T01:22:08.558495+00:00"},{"alias_kind":"pith_short_12","alias_value":"FAMEVBUUXGCE","created_at":"2026-07-17T01:22:08.558495+00:00"},{"alias_kind":"pith_short_16","alias_value":"FAMEVBUUXGCEUWSG","created_at":"2026-07-17T01:22:08.558495+00:00"},{"alias_kind":"pith_short_8","alias_value":"FAMEVBUU","created_at":"2026-07-17T01:22:08.558495+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/FAMEVBUUXGCEUWSGGFC4FRB5BS","json":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS.json","graph_json":"https://pith.science/api/pith-number/FAMEVBUUXGCEUWSGGFC4FRB5BS/graph.json","events_json":"https://pith.science/api/pith-number/FAMEVBUUXGCEUWSGGFC4FRB5BS/events.json","paper":"https://pith.science/paper/FAMEVBUU"},"agent_actions":{"view_html":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS","download_json":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS.json","view_paper":"https://pith.science/paper/FAMEVBUU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.15081&json=true","fetch_graph":"https://pith.science/api/pith-number/FAMEVBUUXGCEUWSGGFC4FRB5BS/graph.json","fetch_events":"https://pith.science/api/pith-number/FAMEVBUUXGCEUWSGGFC4FRB5BS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS/action/storage_attestation","attest_author":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS/action/author_attestation","sign_citation":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS/action/citation_signature","submit_replication":"https://pith.science/pith/FAMEVBUUXGCEUWSGGFC4FRB5BS/action/replication_record"}},"created_at":"2026-07-17T01:22:08.558495+00:00","updated_at":"2026-07-17T01:22:08.558495+00:00"}