Noma Blog

Personal AI Agent Security: Governing dots, Grok Bot, Muse & Muse Code

Personal AI agents like dots, Grok Bot, Instinct, Muse and Muse Code inherit user privileges, operate persistently across workspace channels, and bypass standard security tools and security team reviews. Attacks and flaws, such as agent-to-agent privilege escalation, token theft, and context injection, demonstrate how adversaries can hijack an agent's broad access without breaking system permissions. Noma provides centralized discovery, maps agent blast radiuses via an Agentic Risk Map, and gives security teams visibility and governance over every agent in the enterprise.