OSS-CRS: Liberating AIxCC Cyber Reasoning Systems for Real-World Open-Source Security
summary
An open, locally deployable framework that ports components from all seven AIxCC finalist CRSs and runs them modularly on OSS-Fuzz projects, using a three-stage prepare/build/run pipeline with flat Docker containers and per-CRS CPU/memory limits. It includes an internal LLM proxy for budget management and model redirection, supports seed sharing and ensemble runs across different CRSs, and reproduces Atlantis workflows to find 7 previously unknown bugs (one high-severity) across 6 OSS-Fuzz projects using under $50 of LLM credits per bug.
AI-written summary · about
chapters (10)
- 0:00OSS-CRS and AIxCC intro
- 3:02AIxCC CRS challenges
- 4:04OSS-CRS modular framework
- 5:50OSS-CRS pipeline
- 7:18LLM controls in OSS-CRS
- 8:24Parallel CRS composition
- 10:30CRS interface helpers
- 11:15Proof of life case study
- 13:02AIxCC finalists integration
- 14:06Call to action
related talks
11mAIxCC Results and New Open Source AI Projects To Help Secure Open Sou...
44mSecurity Research with Open Source Software: Vol. Omega
44mArtificial Intelligence Cyber Challenge (AIxCC): Overview and Releasing Research as Open Source Software
53mCracking the Code: Unveiling Synergies Between Open Source Security and AI
01h 02mPanel Discussion:The Present and Future of AI and Security
13mSecuring the Software Commons: Standards, Automation, and AI for a Resilie...
05mLightning Talk: AI Cyber Challenge
21mNavigating the Intersection: AI’s Role in Shaping the Secure Open Source Software Ecos...