AI in the Company: Security and Rules
Rolling language models out across a team: what happens to the data, which rules people actually follow, and how much risk you are choosing to carry
About this course
A course for the person who has to answer for the rollout rather than for their own output: a department head, an IT or security lead, an operations or HR manager, a founder in a company small enough that the question lands on their desk anyway. The subject is not what a model can do, but what happens around it — where the data goes, who holds a copy, which rules survive contact with a deadline, and what each safeguard costs.
Eleven lessons build a working set of artefacts: a three-bucket data classification, a one-page usage policy written in examples instead of prohibitions, an employee memo, a vendor checklist, an agent permission matrix and an incident record. All of them sit in the lessons as templates you can copy and adapt in an afternoon.
The tone is deliberately unalarmed. There is no claim that models steal secrets and no promise that some setting makes a service safe. Risk here is something you measure, reduce to a level you can defend, and then live with. The course also says plainly what most guidance avoids: a blanket ban does not produce zero use, it produces invisible use — and invisible use is the worse outcome. Nothing is tied to a particular product, model version or plan name, and there are no references to the law of any one country: the categories and the practices travel, the statutes do not.