A policy nobody can enforce is a document, not a control.
Every business we talk to has AI in use and most have no logging. The exposure is not the technology. It is that nobody can say what left the building or when.
What is the actual risk of consumer AI tools?
Client information leaving your control with no record. If a client asks whether their file was used to train a model, you need an answer.
Can we just block the sites?
Blocking pushes the work to personal phones and personal accounts. A sanctioned option with logging is easier to defend.
What does an auditor want to see?
An approved tool list, a policy, evidence the policy is enforced technically, and logs. In that order.
Where does a private AI desktop fit?
When the work involves sensitive files, a hosted desktop with a governed AI assistant keeps the data and the audit trail inside Canada.
