AI-Powered DeFi Hacking: Anthropic Research Shows Profitable Autonomous Exploitation Now Feasible

This is a versioned snapshot of the AI-Powered DeFi Hacking post as of 26 March 2026. The current version is at russellclare.com/ai-defi-autonomous-exploitation/.

Anthropic AI agents autonomously scanned 2,849 deployed smart contracts, found 2 novel vulnerabilities, and produced $3,694 in exploits while spending only $3,476 in compute costs. Initial publish.