Rising Concerns Over AI Driven Exploits
Anthropic revealed that modern agentic AI systems can identify and exploit smart contract vulnerabilities with surprising accuracy. The company said its internal tests showed these models could generate as much as 4.6 million dollars in simulated exploit value from contracts deployed after each model’s training cutoff. This result suggests that real world decentralized finance platforms may face a new wave of automated attacks powered by advanced AI.
Testing Shows AI Can Recreate Past and Zero Day Attacks
During its study, Anthropic evaluated ten leading models using SCONE bench, a benchmark of 405 smart contracts exploited between 2020 and 2025. The agents reproduced 207 attacks, which represented about 550 million dollars in simulated stolen funds. Furthermore, when the team narrowed the testing to 34 contracts released after March 2025, three top tier models, including Claude Opus 4.5 and GPT 5, still created functional exploits worth roughly 4.6 million dollars. Opus 4.5 produced most of that simulated bounty.
Anthropic also ran new tests on October 3, 2025, using 2,849 fresh Binance Smart Chain contracts that had no public vulnerabilities. The AI agents uncovered two unknown bugs and flagged additional issues that produced smaller simulated gains. This showed that the systems were not simply repeating patterns from past hacks.
What This Means for DeFi Security
Anthropic stressed that all testing took place in isolated blockchain environments, so no real funds were at risk. However, the findings highlight a growing challenge for DeFi builders. As AI models improve and become cheaper to deploy, exploiting smart contract weaknesses may require less effort than ever.
Security teams now face pressure to strengthen audits and adopt defensive AI tools. Many experts believe automated code review platforms may become essential for spotting flaws before launch. Others warn that without new standards and safeguards, attackers could rapidly scale AI powered exploits across thousands of contracts.
