Satya Nadella says to treat AI models as insider risks. He's right.
The line worth keeping: "We simply can't outsource responsibility for what intelligence does on our behalf."
His answer is the playbook companies already use for powerful insiders. Establish identity. Limit privileges. Log activity. Build containment. Keep a hand on the emergency brake.
Here is the question that goes unasked.
Is that playbook working for your people today?
I've seen more companies than I can count on both hands with a shared copier where people store private information. Someone scans a document and it saves to a common location. Nobody ever cleans it up. Financial records. Banking records. Other people in the company pick it up, and it spreads.
That's an insider risk with no AI in the building.
In more than forty years I have watched the same gap in ERP, in manufacturing, in every kind of organizational change. A commitment gets made before the organization has the capability to keep it.
Containing a model takes an organization that can contain a copier.
That can be measured before the contract is signed.
Unready doesn't equal unbuildable.
Dominic Luzi · Chief Answer Grape, Management Alchemy
