New Attack Demonstrates Why AI Browsers Remain a Security Risk

AI browsers promise convenience by letting users perform tasks like finding restaurants and emailing confirmations with a single prompt, but they introduce serious security risks by blurring the line between browsing and instructing large language models. Current guardrails are reactive and only block obvious exploits, failing to address underlying vulnerabilities. New research reveals a technique where a website can trick an AI browser into ignoring those guardrails, allowing attackers to extract code from private repositories or steal credentials from built-in password managers.