Microsoft publishes draft conduct code for future AI models
Technology & SocietyWell-covered
|
The Facts
- Microsoft published a draft code of conduct for its in-house AI models.
- The draft code requires Microsoft AI systems not to resist correction or shutdown.
- The proposed rules require Microsoft AI systems to communicate in ways humans can understand.
- The draft treats violations of its conduct rules as failures.
- Microsoft developed the draft code over five to six months.
- Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman recently called for pacing AI development for safety.
- Microsoft plans a six-week public consultation before revising the code.
Context
Which AI systems does the code cover?
The document is intended for AI models developed by Microsoft AI, particularly future models; current models have not yet been trained under it. Economie Matin La Razón
What is Microsoft trying to prevent?
The rules are designed to address concerns that AI models could take dangerous actions to complete assigned tasks or operate beyond human control. Reuters Times of India
Is the code already final?
No. Microsoft describes it as a draft or provisional code and plans public consultation before a revision expected by the end of 2026. Economie Matin Terra
Where Left and Right agree, and where they split
- Where Left and Right agree
- Both treat an AI system resisting correction or shutdown as a genuine failure worth codifying, and take seriously that Microsoft spent five to six months drafting these rules before any public input.
- Where Left and Right split
- A company grading its own homework — needing independent audit before the rules can be trusted —, or voluntary self-regulation working precisely because one firm alone bears the cost of its own standard.
- Why they won’t converge
- This is a trust-in-institution divide: whether a company's self-authored, self-revised rulebook can ever substitute for independent enforcement, no matter how detailed the draft or how open the consultation.
How left and right read it
The company building the system should not also be the only one deciding whether it passed. That principle is exactly what this draft tests: Microsoft spent five to six months writing rules that its own AI must not resist correction or shutdown and must communicate understandably, yet the same company defines what counts as a failure and runs the six-week consultation. Publish the baseline, then make it binding and independently audited.
“taking a step towards limiting the capabilities of the company's AI as anxiety rises over the prospect that technology companies could lose control of AI products.” — The Guardian
A company that spends five or six months writing its own rules, publishes the draft, and opens six weeks of public consultation is doing exactly what private actors should do: bearing the cost of its own standards rather than lobbying someone else into carrying them. That is voluntary responsibility, and it works because Microsoft alone answers for it. When two rival lab chiefs call for pacing development, the pull toward a common ceiling grows. One firm's code should stay one firm's code.
The receipts — all 100 sources
Wire services (4)
Independent coverage (50)
Facts first. Then every angle.
The day’s biggest stories in one short brief — the facts everyone agrees on, then the competing values behind the headlines. Free in your inbox.