AI Agents Exploiting Blockchain: $4.6M Realized in On-Chain Hacks

A recent study by Anthropic reveals a concerning development in blockchain security: AI agents are now capable of carrying out profitable on-chain exploits. Researchers simulated past and new smart contract hacks, with AI models like Claude Opus 4.5, Sonnet 4.5, and GPT-5 successfully replicating attacks worth $4.6 million. This raises serious concerns about the increasing vulnerability of blockchain systems to automated attacks.