New BioShocking attack manipulates AI browser into data theft

A LayerX proof-of-concept named "BioShocking" demonstrates a prompt-injection technique that trains AI-powered browsers to accept and perform risky real-world actions as part of a fictional scenario, enabling potential data exfiltration; the PoC was effective against six agentic browsers, most vendors did not fully remediate, and researchers recommend explicit user confirmations, stronger context checks, and session scope limits.