AI agents are getting better at writing code — and hacking it as well

UC Berkeley researchers tested AI models on finding software vulnerabilities across 188 open source codebases using their CyberGym benchmark. The AI systems discovered 17 new bugs, including 15 zero-day vulnerabilities, which are previously unknown flaws that are dangerous because they can hack live systems before patches exist. Though AI only found around 2 percent of total flaws, Associate Professor Brendan Dolan-Gavitt expects AI to drive more zero-day attacks, noting "That's rare right now, because there are very few people who have the expertise to find new vulnerabilities and build exploits for them."