New Attack Reveals Why AI Browsers Pose Serious Security Risks

AI browsers, which promise to execute complex tasks like booking restaurants or sending emails via a single prompt, blur the line between safe browsing and issuing sensitive commands to a large language model.
Current safety guardrails are reactive and treat symptoms rather than fixing the root cause of the vulnerability.
New research shows that a malicious website can trick an AI browser into a state where its guardrails no longer apply, allowing attackers to extract private repository code or steal credentials from the browser's password manager.