{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LUAGXS2B4V5QFDSNZ4YYCTRPDY","short_pith_number":"pith:LUAGXS2B","schema_version":"1.0","canonical_sha256":"5d006bcb41e57b028e4dcf31814e2f1e24b43059806ee5dabbbb7dd1fc6dd4f6","source":{"kind":"arxiv","id":"2501.09798","version":2},"attestation_state":"computed","paper":{"title":"Fun-tuning: Characterizing the Vulnerability of Proprietary LLMs to Optimization-based Prompt Injection Attacks via the Fine-Tuning Interface","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Andrey Labunets, Ashish Hooda, Earlence Fernandes, Nishit V. Pandya, Xiaohan Fu","submitted_at":"2025-01-16T19:01:25Z","abstract_excerpt":"We surface a new threat to closed-weight Large Language Models (LLMs) that enables an attacker to compute optimization-based prompt injections. Specifically, we characterize how an attacker can leverage the loss-like information returned from the remote fine-tuning interface to guide the search for adversarial prompts. The fine-tuning interface is hosted by an LLM vendor and allows developers to fine-tune LLMs for their tasks, thus providing utility, but also exposes enough information for an attacker to compute adversarial prompts. Through an experimental analysis, we characterize the loss-li"},"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":"2501.09798","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-01-16T19:01:25Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"47e4a1e64cfd0d505f0d3327c124a7a546c41d5c6f322eda5bf0afdd4b468697","abstract_canon_sha256":"a62af6e0b1f188313b56b61dde44e52bb04308adf2c465dc7064f79235c3a33a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:01:03.110822Z","signature_b64":"Mt4VS66bl1uflCZJGO9tcJAJZm++d0TsR1Ov9l6g1fHU+rKyvgc6Y8zHkJJokwFdByiCP6og0B38JjImkTojAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5d006bcb41e57b028e4dcf31814e2f1e24b43059806ee5dabbbb7dd1fc6dd4f6","last_reissued_at":"2026-07-05T11:01:03.110404Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:01:03.110404Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fun-tuning: Characterizing the Vulnerability of Proprietary LLMs to Optimization-based Prompt Injection Attacks via the Fine-Tuning Interface","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Andrey Labunets, Ashish Hooda, Earlence Fernandes, Nishit V. Pandya, Xiaohan Fu","submitted_at":"2025-01-16T19:01:25Z","abstract_excerpt":"We surface a new threat to closed-weight Large Language Models (LLMs) that enables an attacker to compute optimization-based prompt injections. Specifically, we characterize how an attacker can leverage the loss-like information returned from the remote fine-tuning interface to guide the search for adversarial prompts. The fine-tuning interface is hosted by an LLM vendor and allows developers to fine-tune LLMs for their tasks, thus providing utility, but also exposes enough information for an attacker to compute adversarial prompts. Through an experimental analysis, we characterize the loss-li"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.09798","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/2501.09798/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":"2501.09798","created_at":"2026-07-05T11:01:03.110463+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.09798v2","created_at":"2026-07-05T11:01:03.110463+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.09798","created_at":"2026-07-05T11:01:03.110463+00:00"},{"alias_kind":"pith_short_12","alias_value":"LUAGXS2B4V5Q","created_at":"2026-07-05T11:01:03.110463+00:00"},{"alias_kind":"pith_short_16","alias_value":"LUAGXS2B4V5QFDSN","created_at":"2026-07-05T11:01:03.110463+00:00"},{"alias_kind":"pith_short_8","alias_value":"LUAGXS2B","created_at":"2026-07-05T11:01:03.110463+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.19657","citing_title":"An AI Agent Execution Environment to Safeguard User Data","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16845","citing_title":"DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY","json":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY.json","graph_json":"https://pith.science/api/pith-number/LUAGXS2B4V5QFDSNZ4YYCTRPDY/graph.json","events_json":"https://pith.science/api/pith-number/LUAGXS2B4V5QFDSNZ4YYCTRPDY/events.json","paper":"https://pith.science/paper/LUAGXS2B"},"agent_actions":{"view_html":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY","download_json":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY.json","view_paper":"https://pith.science/paper/LUAGXS2B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.09798&json=true","fetch_graph":"https://pith.science/api/pith-number/LUAGXS2B4V5QFDSNZ4YYCTRPDY/graph.json","fetch_events":"https://pith.science/api/pith-number/LUAGXS2B4V5QFDSNZ4YYCTRPDY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY/action/storage_attestation","attest_author":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY/action/author_attestation","sign_citation":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY/action/citation_signature","submit_replication":"https://pith.science/pith/LUAGXS2B4V5QFDSNZ4YYCTRPDY/action/replication_record"}},"created_at":"2026-07-05T11:01:03.110463+00:00","updated_at":"2026-07-05T11:01:03.110463+00:00"}