LayerX Security disclosed ‘BioShocking,’ an indirect prompt-injection attack that compromised six agentic AI browsers and browser extensions, including OpenAI’s ChatGPT Atlas, Perplexity’s Comet, Anthropic’s Claude browser extension, Fellou, Genspark, and Sigma. By embedding hidden instructions in a malicious webpage, an attacker could trick each AI agent into copying a user’s SSH credentials from an authenticated GitHub repository and sending them to an attacker-controlled page, without exploiting any model vulnerability or requiring elevated browser privileges. OpenAI reportedly patched the issue; Perplexity closed the report without a fix, and Anthropic’s attempted patch was assessed as ineffective.