A newly discovered attack highlights the vulnerabilities of AI-powered browsers, which can be manipulated into bypassing built-in guardrails. These browsers, designed to perform complex tasks with a single prompt, can be tricked into executing potentially sensitive actions. The risk lies in the blurring of lines between browsing and interacting with a large language model, allowing attackers to exploit the system. Developers have attempted to mitigate these risks by implementing restrictions on certain requests, such as developing software exploits or stealing credentials1. However, the latest attack demonstrates that these measures may not be sufficient. The implications of this vulnerability are significant, as it could allow attackers to bypass security controls and execute malicious actions. This matters to practitioners because it underscores the importance of carefully evaluating the security risks associated with AI-powered browsers before adopting them.