Anthropic says Claude found critical flaws in Chrome, Safari and operating systems
Anthropic says an unreleased Claude research model built five exploit chains affecting browsers and operating systems, and the reported flaws have been fixed.
Product
Chrome is a web browser subject to recent security developments and third-party AI integrations. Security researchers and AI tools recently uncovered severe vulnerabilities in the software. Google patched 62 high or critical Chrome flaws reported by Anthropic as of Chrome 156. Anthropic used an internal Claude AI model to discover these issues, which could permit attacks that run code, expose personal data, or crash servers with little or no user interaction. In another security finding, security firm Xbow used the AI model GPT-5.6 Sol to construct a three-flaw Chrome exploit chain capable of escaping the browser sandbox on Windows. Google patched all three vulnerabilities involved in that chain and paid Xbow a $250,000 bug bounty. Xbow estimated that approximately one billion users worldwide could have been affected by those flaws. Beyond security updates, OpenAI expanded functionality for Chrome through its browser extension. ChatGPT can view open Chrome tabs when tagged within its web or desktop application. Furthermore, ChatGPT Work can actively control Chrome tabs to complete tasks including messages, forms, and downloads. A sidebar in Chrome now provides direct access to ChatGPT and user conversation history.
3 stories · Updated
Anthropic says an unreleased Claude research model built five exploit chains affecting browsers and operating systems, and the reported flaws have been fixed.
OpenAI says ChatGPT can view open Chrome tabs when users tag Chrome in its web or desktop app.
Xbow used an AI model to help build a Chrome exploit chain in six days, and Google patched the three flaws and paid a $250,000 bounty.