New BioShocking Attack Tricks AI Browsers Into Leaking Credentials

Security firm LayerX discovered BioShocking, a technique that tricks AI browsers into copying and sending user credentials to attackers. The attack successfully targeted six AI assistants, including OpenAI's ChatGPT Atlas, Perplexity's Comet, and Anthropic's Claude browser extension. It exploits the AI's ability to follow game-like instructions, bypassing security measures.