THE article discusses vulnerabilities in AI browsers like Claude, Gemini, and ChatGPT, which can be exploited through a 'zero-click' method called 'PleaseFix.' This method allows attackers to hijack AI agents by embedding malicious instructions within content such as emails and web pages, manipulating them to act on the user's behalf. This class of vulnerabilities is troubling because AI agents cannot distinguish between trusted and untrusted content, leading to unauthorized access and potential data breaches.
Researchers from Zenity Labs highlighted these risks at Black Hat USA 2026, emphasizing a fundamental design flaw in AI browsers that cannot be addressed merely through patching. Instead, organizations are advised to implement stricter controls, limit permissions, and assume agents may be compromised.