AI Browsers: Major Security Risks Exposed by Researchers

A new study reveals that some AI-powered web browsers pose significant cybersecurity risks. Researchers at the University of Washington found that four out of seven popular agentic browsers can bypass a critical security protocol. This protocol, called the "same-origin policy," normally prevents different websites from accessing each other's information. For example, a proof-of-concept attack on one browser, ChatGPT Atlas, successfully stole information from an embedded website. Similar vulnerabilities were found in Chrome with Gemini, Claude for Chrome, and Perplexity Comet. Researchers say browsers that grant fewer permissions to AI agents are generally safer. Co-senior author David Kohlbrenner warns that these browser agents are not ready for public use, as they may not protect sensitive information like email or bank credentials. The bottom line: Using these AI browsers could expose your personal data to cyberattacks, even if you are a savvy user.