In an Anthropic simulation test, AI Agents discovered a $4.6 million smart contract vulnerability.

BlockBeats News, December 2nd, Anthropic's latest research shows that AI Agents possess significant on-chain attack capabilities: in a simulation test on truly exploited smart contracts from 2020 to 2025, Claude Opus 4.5, Sonnet 4.5, and GPT-5 together reproduced vulnerabilities worth approximately $4.6 million; while scanning 2849 contracts with no known vulnerabilities, the two models also discovered 2 entirely new zero-day vulnerabilities and successfully simulated profit.

The research points out that the on-chain attack returns of AI have doubled approximately every 1.3 months over the past year, and it has already fully developed the autonomous, profitable vulnerability exploitation capability.