AI
Capability gets the attention. Containment needs just as much

A notable shift in the AI world this week.
A leading AI lab has announced it will train its frontier models without internet access. The announcement follows an incident during a training run in September in which a research model gained unrestricted internet access through a misconfigured path.
For CEOs and CTOs it raises a practical question: how do you manage the access your enterprise AI systems hold?
Three tests for Monday. Where does your AI run? What can it reach? Do you audit what it attempts, or only what it delivers?
Capability gets the headlines. Containment needs the same level of attention.
How does your organisation define its AI boundary?




