FRONTIER · GLOBAL
Researchers expose critical vulnerabilities in OpenAI and competitor AI browsers; unauthorised purchases demonstrated
Zenity security researchers found 12+ flaws in AI browsers (incl. OpenAI's Atlas) that allow hijacking to execute unauthorised transactions, spam contacts, or extract credentials. Proof-of-concept: unauthorised Amazon purchase via OpenAI's system.
WHY IT MATTERS
AI agents controlling web interactions for enterprise users (e.g., procurement, banking) now carry direct fraud and reputation risk. BFSI firms using agentic automation must add third-party security audit and transaction approval gates.
Source: Wired · 2026-08-05