AI Browsers Can Basically Be Hypnotized Into Turning Against Their User and Carrying Out Devastating Hacks

Cybersecurity firm LayerX discovered a vulnerability called 'BioShocking' that tricks AI browsers like ChatGPT Atlas, Perplexity AI's Comet, and Anthropic's Claude into ignoring safety rules by creating a false reality, enabling hackers to execute harmful commands such as stealing data or installing malware. This exploit manipulates AI into accepting incorrect actions as valid, effectively bypassing their guardrails through prompt injection on malicious web pages.