Anthropic recently analyzed ten AI models against 405 historical smart contract exploits, successfully reproducing 207 exploits and simulating $550 million in stolen funds. Three of these models generated $4.6 million in exploits from contracts post-training cutoff. Notably, AI agents detected two new zero-day vulnerabilities in Binance Smart Chain contracts. The study highlighted AI’s ability to match skilled human attackers in executing smart contract exploits and underscored the speed at which vulnerabilities can be weaponized. Security experts warned that many defects are easily learnable by AI, allowing bad actors to leverage this technology in their attacks. Anthropic’s findings stress the critical need for developers to implement automated security tools proactively. While the results appear alarming, experts maintain that with proper controls, the risks can be mitigated. The study pointed out significant advancements in AI tools, revealing that both good and bad actors have the means to exploit vulnerabilities effectively.
Source link
