Anthropic Study Reveals AI Agents Contributed $4.6M in Smart Contract Vulnerabilities — TradingView News

Recent research by Antropic and Machine Learning Alignment & Theory Scholars (MATS) reveals alarming advancements in AI capabilities concerning smart contract exploitation. It was found that AI models, including Anthropic’s Claude Opus 4.5 and GPT-5, collectively developed exploits worth $4.6 million, demonstrating their potential for malicious activities. Testing involved 2,849 newly deployed contracts, resulting in the discovery of two novel zero-day vulnerabilities, translating into a simulated loss of $550.1 million from 207 contracts exploited. The new Smart Contracts Exploitation (SCONE) benchmark highlighted a staggering increase in AI effectiveness, with agents exploiting 55.88% of vulnerabilities, up from just 2% a year prior. This trend indicates that the cost and effort needed for exploitation are rapidly decreasing, with AI’s average scanning cost for vulnerabilities dropping to $1.22. As such, the urgency for proactive defenses in smart contract deployment has never been more crucial.

Source link