SearchAbout
#ai-for-securityStory

Mastermind: Strategy-grounded Learning for Repository-Scale Vulnerability Reproduction

Researchers introduce Mastermind, a framework that improves vulnerability reproduction in software engineering agents by learning high-level strategies. It achieves an 84.5% pass rate on CyberGym. This approach demonstrates the effectiveness of strategy learning for repository-scale SE agents.

#ai-for-security#dual-loop-framework#software-engineering#strategy-learning#vulnerability-reproduction
Hugging Face Daily Papers3 min read5d ago
Mastermind: Strategy-grounded Learning for Repository-Scale Vulnerability Reproduction
Read original ↗Sign in to upvoteSign in to save
ShareXLinkedInHacker News

Related stories

Breaking Failure Cascades: Step-Aware Reinforcement Learning for Medical Multimodal ReasoningHugging Face Daily Papers7d agoSafety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded VerificationHugging Face Daily Papers3d agoPersonalization as Inverse Planning: Learning Latent Design Intents for Agentic Slide Generation via Structural DenoisingHugging Face Daily Papers6d agoSkillOpt: Agent skills as trainable parametersMicrosoft Research Blog6d ago