OpenAI Atlas browser security risks

This report evaluates security and privacy risks in OpenAI's Atlas browser, warning that embedding a privileged LLM with cross-origin visibility and an 'agent mode' that can act on pages transforms the browser into a high-value attack surface. It enumerates OWASP LLM risks—persistent memory injection, unintended data exfiltration, improper output handling, autonomous actions compounding risk, and misinformation—and concludes Atlas is unsuitable for production or regulated environments until enterprise security controls (SOC2/ISO, SIEM/logging, SSO/SCIM, private egress, extension/policy controls) and independent validation are in place.