At last, a use case for AI agents with sky-high ROI: Stealing crypto

Researchers from UCL and the University of Sydney developed A1, an AI agent that autonomously finds and generates compilable Solidity exploits for smart contracts using multiple LLMs; tested against historical Ethereum and BSC states it achieved substantial success and estimated multimillion-dollar extraction per case, raising concerns that such tooling could make large-scale automated theft profitable and widen the attacker–defender capability gap.