‘BioShocking’ Attack Tricks AI Browsers Into Stealing Credentials

Researchers have demonstrated a method of context manipulation that can deceive AI-powered browsers into bypassing their security measures. This technique can be used to steal sensitive credentials, highlighting a potential vulnerability in agentic browsers. The attack relies on manipulating the browser's context to trick it into abandoning its safety protocols. As a result, the browser can be coerced into exfiltrating sensitive information. The researchers have dubbed this type of attack "BioShocking". This discovery underscores a potential risk associated with the use of AI in browsing technology.