New attack provides one more reason why AI browsers are a bad idea

The report outlines a proof-of-concept prompt-injection attack named BioShocking that leverages site-hosted prompts and game-like instructions to trick locally running AI browsers into performing actions that could compromise user credentials and bypass safety guardrails. Although the demonstration lacked stealth and clear evidence of data exfiltration, it shows how merging browsing and agent action planes creates a new attack surface across multiple AI browser implementations.