AI helped researchers break Chrome in six days, security firm says
Security firm XBOW says its researchers used a public OpenAI model to build a full attack on Google Chrome in six days. The attack chained three bugs to take over a Windows computer from a web page. Google fixed the bugs and paid XBOW $250,000, The Hacker News reported.
Security firm XBOW says its researchers used a publicly available OpenAI model to build a complete attack on Google's Chrome browser in just six days, The Hacker News reported. Google has confirmed and fixed the three bugs and paid XBOW $250,000, its top reward for this kind of report.
The attack is called a full chain. It links several bugs so that simply loading a web page lets an attacker run code on a Windows computer, The Hacker News said. The user would not need to click or download anything.
Chrome keeps web pages inside layers of protection, called sandboxes. This chain broke through both of them, the outlet reported.
XBOW said the bugs were never used against Chrome users. The fixes shipped in Chrome updates on Aug. 25 and Sept. 3, The Hacker News reported. The fix only works after the browser restarts, so users should relaunch Chrome.
XBOW built the winning chain from Aug. 2 to Aug. 7, during hacker conferences in Las Vegas, The Hacker News said. That followed months of earlier work. The company said the six-day run cost about $40,000 in AI charges.
Forbes reported that XBOW engineer Brendan Dolan-Gavitt used OpenAI's model to chain the three flaws together. The outlet said XBOW is a $1 billion AI security startup.
People and AI split the work, The Hacker News reported. The AI agents searched Chrome's code, tested ideas and wrote early drafts of the attack. The researchers picked targets, checked the results and decided which pieces to join.
"Pre-LLMs a Chrome full chain would take a team of four around 6-9 months to be produced, our first submission took 2 months. The awarded one? six days," Dolan-Gavitt wrote, according to The Hacker News.
He also had a warning for defenders. "Six days and $40k is not a lot of time or money when the payoff is a Chrome chain!" he wrote, The Hacker News reported. "Defend accordingly."
What we know: Google fixed three Chrome bugs that XBOW chained with help from a public AI model, and paid a $250,000 reward.
What we don't: How easily other attackers could repeat the work without XBOW's expert researchers.
Where outlets agree The Hacker News and Forbes both report that XBOW used OpenAI's model to chain three Chrome flaws and that Google paid a $250,000 reward.
Where they differ The outlets cited here report the same key facts so far.