AI agents may outnumber humans: security experts say treat each as untrusted identity

CurieuxExplorer · x · 2026-09-30

Dinis Guarda's article examines security governance as AI agents scale. A Forbes Fermi estimate—built from 3M+ models on Hugging Face, billions of downloads, and a cloud multiplier from Anthropic's reported $47B annualized revenue run rate—puts active agents anywhere from 800 million to 8 billion, while formal enterprise deployments tracked by Statista sit around 28.6 million; IDC projects 1.3 billion formal agents by 2028.

The core argument: traditional security models built around human users and static software no longer suffice when agents can access systems, hold credentials, and execute actions autonomously. The starting point: treat every agent as an untrusted identity—scoped credentials, no standing admin rights, and a sanctioned way to stop them.

Original post →

More from AGI Musings

AGI Musings channel →