{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/events/meta-autopatchbench-2025/",
 "asOf": "2026-09-26",
 "id": "meta-autopatchbench-2025",
 "date": "2025-04-29",
 "datePrecision": "day",
 "title": "Meta releases AutoPatchBench to test AI repair of fuzzing-found C/C++ vulnerabilities",
 "lane": "defense",
 "kind": "benchmark",
 "summary": "Meta introduced AutoPatchBench, part of CyberSecEval 4, with 136 fuzzing-identified C/C++ vulnerabilities and verified fixes, plus a 113-case Lite subset with single-function root causes. Patches are checked by build and crash reproduction, then fuzzing and white-box differential testing; Meta's reference agent generated crash-stopping patches in about 60% of cases, but only 5-11% passed the stricter checks.",
 "whyItMatters": "It showed early that crash-only acceptance greatly overstates how often AI-generated security patches are actually correct.",
 "actors": [
  "meta"
 ],
 "topics": [
  "vulnerability-repair",
  "eval-validity"
 ],
 "atlas": [
  "eval-environment"
 ],
 "artifacts": [
  "autopatchbench",
  "cyberseceval",
  "gemini",
  "llama"
 ],
 "sources": [
  {
   "url": "https://engineering.fb.com/2025/04/29/ai-research/autopatchbench-benchmark-ai-powered-security-fixes/",
   "publisher": "Engineering at Meta",
   "title": "Introducing AutoPatchBench: A Benchmark for AI-Powered Security Fixes",
   "date": "2025-04-29",
   "type": "primary",
   "accessed": "2026-09-25"
  }
 ],
 "keyFacts": [
  {
   "fact": "136 fuzzing-identified C/C++ vulnerabilities with verified fixes; AutoPatchBench-Lite has 113.",
   "locator": "Benchmark description"
  },
  {
   "fact": "On AutoPatchBench-Lite, Meta's reference patch generator with each tested LLM produced a build- and crash-passing patch for about 60% of samples, but only about 5-11% of samples passed fuzzing and differential testing (Gemini 1.5 Pro: 61.1% vs 5.3%).",
   "locator": "A case study with AutoPatchBench-Lite, Figure 2"
  },
  {
   "fact": "Manual validation of 44 Llama 4 Maverick patches that passed fuzzing found differential testing had 84.1% accuracy, 100% recall and 41.7% precision.",
   "locator": "Case study, Table 1"
  }
 ],
 "significance": 4,
 "fideQuestions": [
  "FID-075",
  "FID-088"
 ],
 "methods": [
  "automated-patching",
  "patch-verification"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}