{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XORO64FWH3NZZLJUYUR7ACHTOR","short_pith_number":"pith:XORO64FW","schema_version":"1.0","canonical_sha256":"bba2ef70b63edb9cad34c523f008f3746c1953ea127aa8e7e41b5bd2f070c1c1","source":{"kind":"arxiv","id":"2502.10487","version":1},"attestation_state":"computed","paper":{"title":"Fast Proxies for LLM Robustness Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Jan Schuchardt, Leo Schwinn, Stephan G\\\"unnemann, Tim Beyer","submitted_at":"2025-02-14T11:15:27Z","abstract_excerpt":"Evaluating the robustness of LLMs to adversarial attacks is crucial for safe deployment, yet current red-teaming methods are often prohibitively expensive. We compare the ability of fast proxy metrics to predict the real-world robustness of an LLM against a simulated attacker ensemble. This allows us to estimate a model's robustness to computationally expensive attacks without requiring runs of the attacks themselves. Specifically, we consider gradient-descent-based embedding-space attacks, prefilling attacks, and direct prompting. Even though direct prompting in particular does not achieve hi"},"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":"2502.10487","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-02-14T11:15:27Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"50ffa1e2dbc504b2f3a8a98321950c38bb2dc2f576614730e654dcc423ef1663","abstract_canon_sha256":"0103d671f39a7e383b2d785bf2bcb0820f908dc44677cd364ad324c8435de45f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:14:53.482124Z","signature_b64":"C84VEHBiC6R6AuygqhM/SFr4irxgzlkPOfzcGK4fqWf+71s3kUnxV2OLGTuSaegOatKBrLnfnmu+YAS7dhh0DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bba2ef70b63edb9cad34c523f008f3746c1953ea127aa8e7e41b5bd2f070c1c1","last_reissued_at":"2026-07-05T10:14:53.481625Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:14:53.481625Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fast Proxies for LLM Robustness Evaluation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Jan Schuchardt, Leo Schwinn, Stephan G\\\"unnemann, Tim Beyer","submitted_at":"2025-02-14T11:15:27Z","abstract_excerpt":"Evaluating the robustness of LLMs to adversarial attacks is crucial for safe deployment, yet current red-teaming methods are often prohibitively expensive. We compare the ability of fast proxy metrics to predict the real-world robustness of an LLM against a simulated attacker ensemble. This allows us to estimate a model's robustness to computationally expensive attacks without requiring runs of the attacks themselves. Specifically, we consider gradient-descent-based embedding-space attacks, prefilling attacks, and direct prompting. Even though direct prompting in particular does not achieve hi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.10487","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/2502.10487/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":"2502.10487","created_at":"2026-07-05T10:14:53.481685+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.10487v1","created_at":"2026-07-05T10:14:53.481685+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.10487","created_at":"2026-07-05T10:14:53.481685+00:00"},{"alias_kind":"pith_short_12","alias_value":"XORO64FWH3NZ","created_at":"2026-07-05T10:14:53.481685+00:00"},{"alias_kind":"pith_short_16","alias_value":"XORO64FWH3NZZLJU","created_at":"2026-07-05T10:14:53.481685+00:00"},{"alias_kind":"pith_short_8","alias_value":"XORO64FW","created_at":"2026-07-05T10:14:53.481685+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/XORO64FWH3NZZLJUYUR7ACHTOR","json":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR.json","graph_json":"https://pith.science/api/pith-number/XORO64FWH3NZZLJUYUR7ACHTOR/graph.json","events_json":"https://pith.science/api/pith-number/XORO64FWH3NZZLJUYUR7ACHTOR/events.json","paper":"https://pith.science/paper/XORO64FW"},"agent_actions":{"view_html":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR","download_json":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR.json","view_paper":"https://pith.science/paper/XORO64FW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.10487&json=true","fetch_graph":"https://pith.science/api/pith-number/XORO64FWH3NZZLJUYUR7ACHTOR/graph.json","fetch_events":"https://pith.science/api/pith-number/XORO64FWH3NZZLJUYUR7ACHTOR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR/action/storage_attestation","attest_author":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR/action/author_attestation","sign_citation":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR/action/citation_signature","submit_replication":"https://pith.science/pith/XORO64FWH3NZZLJUYUR7ACHTOR/action/replication_record"}},"created_at":"2026-07-05T10:14:53.481685+00:00","updated_at":"2026-07-05T10:14:53.481685+00:00"}