AI browser vulnerabilities are worrying but the industry response may worry you more

A team from the University of Washington's Allen School of Computer Science & Engineering, led by David Kohlbrenner and Franziska Roesner, tested seven agentic AI browsers for weaknesses in the same-origin policy, one of the web's core security protections. Following responsible disclosure practices, the researchers shared their findings with every company before publishing the study, giving each one an opportunity to respond or fix the issues. The feedback was not consistent: Anthropic, which d