{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PA263E2OKI44VCADG3COFJUTOC","short_pith_number":"pith:PA263E2O","schema_version":"1.0","canonical_sha256":"7835ed934e5239ca880336c4e2a6937093c6adba01b19dcb0f00e872c104063f","source":{"kind":"arxiv","id":"2504.18333","version":1},"attestation_state":"computed","paper":{"title":"Adversarial Attacks on LLM-as-a-Judge Systems: Insights from Prompt Injections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Dmitry Namiot, Narek Maloyan","submitted_at":"2025-04-25T13:18:42Z","abstract_excerpt":"LLM as judge systems used to assess text quality code correctness and argument strength are vulnerable to prompt injection attacks. We introduce a framework that separates content author attacks from system prompt attacks and evaluate five models Gemma 3.27B Gemma 3.4B Llama 3.2 3B GPT 4 and Claude 3 Opus on four tasks with various defenses using fifty prompts per condition. Attacks achieved up to seventy three point eight percent success smaller models proved more vulnerable and transferability ranged from fifty point five to sixty two point six percent. Our results contrast with Universal Pr"},"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":"2504.18333","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-04-25T13:18:42Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d4c0fc358f01acf45c749b44a70e2a9db730ebb89fa9d1243cf47950ca992fa8","abstract_canon_sha256":"a7b846174d7d2dc02e488f9b74e261cd44a89ea50d4a1ff719ba264be4a6464f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:54:05.550933Z","signature_b64":"lrvv1K/08DyRR+/dw3OHjyz9My681z3JKhX+3KebYmNRLLYVBhToXqNU8aCoJBYThCahdQU/UC1zyaEuRq3FDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7835ed934e5239ca880336c4e2a6937093c6adba01b19dcb0f00e872c104063f","last_reissued_at":"2026-07-05T10:54:05.550431Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:54:05.550431Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adversarial Attacks on LLM-as-a-Judge Systems: Insights from Prompt Injections","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CR","authors_text":"Dmitry Namiot, Narek Maloyan","submitted_at":"2025-04-25T13:18:42Z","abstract_excerpt":"LLM as judge systems used to assess text quality code correctness and argument strength are vulnerable to prompt injection attacks. We introduce a framework that separates content author attacks from system prompt attacks and evaluate five models Gemma 3.27B Gemma 3.4B Llama 3.2 3B GPT 4 and Claude 3 Opus on four tasks with various defenses using fifty prompts per condition. Attacks achieved up to seventy three point eight percent success smaller models proved more vulnerable and transferability ranged from fifty point five to sixty two point six percent. Our results contrast with Universal Pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.18333","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/2504.18333/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":"2504.18333","created_at":"2026-07-05T10:54:05.550494+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.18333v1","created_at":"2026-07-05T10:54:05.550494+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.18333","created_at":"2026-07-05T10:54:05.550494+00:00"},{"alias_kind":"pith_short_12","alias_value":"PA263E2OKI44","created_at":"2026-07-05T10:54:05.550494+00:00"},{"alias_kind":"pith_short_16","alias_value":"PA263E2OKI44VCAD","created_at":"2026-07-05T10:54:05.550494+00:00"},{"alias_kind":"pith_short_8","alias_value":"PA263E2O","created_at":"2026-07-05T10:54:05.550494+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07539","citing_title":"Prompt Governance? On Governing Technologies Governed by Natural Language","ref_index":219,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC","json":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC.json","graph_json":"https://pith.science/api/pith-number/PA263E2OKI44VCADG3COFJUTOC/graph.json","events_json":"https://pith.science/api/pith-number/PA263E2OKI44VCADG3COFJUTOC/events.json","paper":"https://pith.science/paper/PA263E2O"},"agent_actions":{"view_html":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC","download_json":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC.json","view_paper":"https://pith.science/paper/PA263E2O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.18333&json=true","fetch_graph":"https://pith.science/api/pith-number/PA263E2OKI44VCADG3COFJUTOC/graph.json","fetch_events":"https://pith.science/api/pith-number/PA263E2OKI44VCADG3COFJUTOC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC/action/storage_attestation","attest_author":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC/action/author_attestation","sign_citation":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC/action/citation_signature","submit_replication":"https://pith.science/pith/PA263E2OKI44VCADG3COFJUTOC/action/replication_record"}},"created_at":"2026-07-05T10:54:05.550494+00:00","updated_at":"2026-07-05T10:54:05.550494+00:00"}