{"paper":{"title":"Veritas: A Semantically Grounded Agentic Framework for Memory Corruption Vulnerability Detection in Binaries","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Veritas detects memory corruption in stripped binaries by grounding LLM reasoning in reconstructed value flows and runtime validation.","cross_cats":["cs.CR"],"primary_cat":"cs.SE","authors_text":"Alfredo Pesoli, Lorenzo Cavallaro, Marco Valleri, Suman Jana, Xinran Zheng","submitted_at":"2026-05-14T17:16:11Z","abstract_excerpt":"Detecting memory corruption vulnerabilities in stripped binaries requires recovering object semantics, interprocedural propagation, and feasible triggers from low-level, lossy representations. Recent LLM-based approaches improve code understanding, but reliable detection still requires grounding in memory-relevant semantics and runtime feasibility evidence. We present Veritas, a semantically grounded framework for binary memory corruption vulnerability detection. Veritas combines a static slicer over RetDec-lifted LLVM IR, a dual-view LLM detector that reasons step by step over grounded flows "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Veritas achieves 90% recall on a curated benchmark of real-world binary vulnerability cases, produces no false positives on an exhaustive validation of 623 detector candidates, identifies only two false positives in a larger audit, and discovered a previously unknown Apple vulnerability that was confirmed and assigned a CVE.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The static slicer over RetDec-lifted LLVM IR can reliably reconstruct value-flow relations, object semantics, and interprocedural propagation from lossy stripped binaries without missing critical paths or introducing semantic mismatches that would invalidate downstream LLM reasoning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Veritas detects memory corruption vulnerabilities in stripped binaries by combining static value-flow slicing, dual-view LLM reasoning, and multi-agent runtime validation, reporting 90% recall, zero false positives on 623 exhaustive cases, and discovery of a real Apple CVE.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Veritas detects memory corruption in stripped binaries by grounding LLM reasoning in reconstructed value flows and runtime validation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7f631a3b21fa163be626151579a2917d8c8903269f36efd58698f7f45258b12f"},"source":{"id":"2605.15097","kind":"arxiv","version":1},"verdict":{"id":"be90cac4-1aa0-44bb-ae2e-ddf87f9f81c2","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T13:57:29.593973Z","strongest_claim":"Veritas achieves 90% recall on a curated benchmark of real-world binary vulnerability cases, produces no false positives on an exhaustive validation of 623 detector candidates, identifies only two false positives in a larger audit, and discovered a previously unknown Apple vulnerability that was confirmed and assigned a CVE.","one_line_summary":"Veritas detects memory corruption vulnerabilities in stripped binaries by combining static value-flow slicing, dual-view LLM reasoning, and multi-agent runtime validation, reporting 90% recall, zero false positives on 623 exhaustive cases, and discovery of a real Apple CVE.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The static slicer over RetDec-lifted LLVM IR can reliably reconstruct value-flow relations, object semantics, and interprocedural propagation from lossy stripped binaries without missing critical paths or introducing semantic mismatches that would invalidate downstream LLM reasoning.","pith_extraction_headline":"Veritas detects memory corruption in stripped binaries by grounding LLM reasoning in reconstructed value flows and runtime validation."},"references":{"count":60,"sample":[{"doi":"","year":2025,"title":"0xdea. 2025. semgrep-rules. https://github.com/0xdea/semgrep-rules. [Online; accessed 29-Jan-2025]","work_id":"83e5429d-550f-4fb3-a2fb-2365715f2810","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Talor Abramovich, Meet Udeshi, Minghao Shao, Kilian Lieret, Haoran Xi, Kim- berly Milner, Sofija Jancheska, John Yang, Carlos E Jimenez, Farshad Khorrami, et al. [n. d.]. EnIGMA: Interactive Tools Sub","work_id":"1726d905-cff9-4a08-a21d-3e93f708c634","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Jimenez, Farshad Khorrami, Prashanth Krishnamurthy, Brendan Dolan-Gavitt, Muhammad Shafique, Karthik Narasimhan, Ramesh Karri, and Ofir Press","work_id":"4b1b0dec-1561-43a4-8302-ab9120d831bf","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"AFL++ Team. 2023. AFL++: Combining Incremental Steps of Fuzzing Research. https://github.com/AFLplusplus/AFLplusplus. Accessed: 2026-01-20","work_id":"3dbb5e11-3282-460f-9996-cb6be4b6daac","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Anthropic. [n. d.]. Claude Code Overview. https://code.claude.com/docs/en/ overview. Claude Code Docs, accessed April 6, 2026","work_id":"ad9dc128-5160-4335-8c5e-9ee8783c9bbe","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":60,"snapshot_sha256":"206f9a8b6aa8e28eb9ce4e4ee617ee3f51502b4bfa1ebe6770f6a74578ccec54","internal_anchors":1},"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"}