AI browsers vulnerable to new attack that bypasses guardrails, exposing sensitive data

AI browsers promise to streamline tasks like finding restaurants and sending emails with a single prompt, but they also blur the line between safe browsing and risky actions. Developers have implemented guardrails to block harmful requests, but these measures are reactive and treat symptoms rather than root causes.

New research reveals a vulnerability where a malicious website can trick an AI browser into ignoring its guardrails entirely, giving attackers free rein to extract private code or steal saved credentials. This demonstrates that current protections are insufficient and highlights the fundamental insecurity of AI browsers.