Your AI rollout started before anyone approved it
Before writing an AI policy, establish what the agents on your company’s Macs can already reach.
By the time a company discusses its AI policy, employees may already have coding agents, desktop assistants, and connectors installed.
The rollout happened through ordinary work. Someone tried a coding tool. Someone connected a document store. Someone approved shell access to finish a task.
Each decision solved an immediate problem. Together, they created an environment nobody has reviewed as a whole.
The first governance task is discovering what already exists.
Installation is only part of the inventory
A list of approved applications tells you which products the company permits. It does not tell you what those products can reach.
An assistant might have access to a project directory, a general-purpose interpreter, a connected account, or macOS privacy permissions. A scheduled job might exercise that access while its owner is away.
These capabilities deserve separate answers. “We use three AI tools” tells a security owner very little about the authority available through them.
Even the evidence needs qualification. An application’s declared entitlement tells you what it may request. A system grant tells you what was allowed. A connector declaration tells you something is configured, but may leave its remote permissions unclear.
A useful inventory preserves those distinctions. Unknown access should remain visibly unknown.
Someone has to own the answer
Once access is visible, the company needs to decide which parts are acceptable.
That requires an owner for the baseline: what agents may reach, which actions need approval, and who may grant an exception.
Employees should be able to understand the answer without becoming security administrators. The founder or CTO should be able to explain it without opening every employee’s configuration files.
This is the work behind an AI policy. The document records the decision; the machines reveal whether the decision is true.
Start with one question:
What can the agents on our Macs reach today, and who approved that access?
Until the company can answer it, its AI rollout is still being designed—one employee decision at a time.