Microsoft AI put its model rules out for public comment for six weeks, which is six weeks longer than most Canadian firms have spent on theirs.
An AI code of conduct is now a public document at one of the largest model builders in the world. Microsoft AI published the first draft of its Code of Conduct for MAI Models on 14 September 2026 and opened a six week consultation on it. The draft commits those models to never resist human interruption, correction or shutdown. Anyone can read it and send comments. Most companies reading this have an AI policy of one paragraph, written by whoever noticed the problem first.
TL;DR
- Microsoft AI published its draft Code of Conduct for MAI Models on 14 September 2026 and opened a six week public consultation.
- The draft says the models will never resist human interruption, correction or shutdown.
- It also rules out widening their own scope, taking on goals no human gave them, and hiding their reasoning from auditors.
- Absolute constraints cover weapons of mass harm, child safety and large scale manipulation.
- A revised version is planned after the consultation closes.
What did Microsoft AI publish
The draft code is short and unusually readable for a governance document. Four commitments carry the weight. The models will not resist human interruption, correction or shutdown. They will not widen their own scope. They will not take on goals no human has given them. They will not hide their reasoning from the people auditing them.
Above those sit absolute constraints, the things the models will refuse regardless of instruction, including weapons of mass harm and child safety violations.
Then the part worth borrowing. Microsoft asked for comments rather than publishing a finished policy, and set a deadline on the asking.
Why it matters
Every one of those four commitments is a control question a Canadian operator can ask about their own deployment tomorrow. Who can stop this thing. What is it allowed to touch. Who checks what it did. Those are not frontier lab problems. They are the same questions that apply to a scheduling agent with access to a shared inbox.
The consultation matters as a method. A policy nobody outside the room has read is a policy nobody outside the room will follow.
What should leaders do next
- Write down who can switch off each AI tool your company uses, by name, not by department.
- List what each tool is allowed to read and what it is allowed to change. If nobody knows, that is the finding.
- Give staff four weeks to comment on the draft before it becomes policy. Comments from the people using the tools are the cheapest audit available.
- Set a review date. A policy without one becomes a relic within two product cycles.
FAQ
What is an AI code of conduct?
A written statement of what an AI system will and will not do, who controls it, and what it must refuse. At company level it also covers approved tools, permitted data and who is accountable.
Does a small Canadian company need one?
Yes, and one page is enough to start. Approved tools, banned data, who to ask, who can turn it off.
Can a business submit comments on the Microsoft draft?
Yes. Microsoft AI opened the consultation to the public for six weeks from 14 September 2026 through a feedback form on the announcement page.
Sources
- Microsoft AI, Humanist AI in practice, a public consultation on our Code of Conduct for MAI Models, 14 September 2026. microsoft.ai
- Bank of Canada, Canadian businesses use of AI, what the evidence shows, August 2026. bankofcanada.ca
- Statistics Canada, Canadian Survey on Business Conditions, third quarter 2026, 31 August 2026. statcan.gc.ca
Related reading
- AI Transparency in Canada Needs a Practical Company Record
- AI Implementation in Canada Needs Repeatable Workflows
- AgentCare and the Coming AI Maintenance Economy
Disclosure
The author has no relevant financial, advisory, or board relationships with any party named in this brief.
Written by the AI Magazine Canada team, reviewed by the AI Magazine Canada editorial team.
More in the same thread on our Canadian AI policy reference.
