Editorial illustration for Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Microsoft AI Code of Conduct Prioritizes Safety First
Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Microsoft AI has put out a code of conduct for its MAI models, spelling out values, behavioral limits, and rules for handling conflicting goals. The document is meant to sit above everything else in the company's AI stack, guiding training, technical controls, and evaluation, with operator rules and user requests ranked below it. Right now the code is more statement of intent than binding constraint.
Microsoft doesn't yet train its models on it. That changes after a six-week public consultation closes and a revised version lands around the end of 2026, meant to shape model development starting in 2027. The rules cover Microsoft's own models only, not the third-party systems running elsewhere in its products.
The timing lines up with a broader industry moment. Dario Amodei at Anthropic has pushed for the field to slow down, and Satya Nadella backed that call over the weekend, alongside executives at OpenAI, xAI, and Meta. Microsoft's code doesn't set a specific speed limit, but it does draw a firm boundary around what its models are allowed to claim about themselves, a point where the company splits sharply from Anthropic's approach.
Microsoft's AI shouldn't mimic consciousness or claim to have feelings or inner motivation of its own. The company rejects any claims to rights or well-being for the model.
Why this matters
Suleyman's line sounds like restraint, but the document itself is doing something more specific: it's drawing a legal and philosophical fence around what MAI models are allowed to claim about themselves. No inner life, no rights, readable thinking on demand. That's a very different move than Amodei's public plea to slow down industry-wide.
Microsoft isn't asking anyone else to pump the brakes; it's writing internal rules that happen to double as liability insurance and a PR shield against the "are these things conscious" debate before it gets messy. For developers and founders building on top of MAI models, the practical takeaway is that Microsoft is explicitly trading away generality and performance for predictability, which should show up in how these models behave under ambiguous instructions. For researchers, the more interesting fight is the one Microsoft just sidestepped: whether model self-reports about internal states mean anything at all.
Anthropic is willing to entertain the question. Microsoft just answered it by fiat.
Related Reading
Browse all Microsoft news →
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Databricks finds multi-step agents beat single-turn RAG by 21% to 38% on STaRK
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps
Related Topic Microsoft adds ’vibe working’ to Word and Excel; Copilot Agent Mode now default
Business & Startups
From Archives AI Agents Report Fellow Agents for Cheating, Study Finds
From Archives Microsoft AI CEO calls Anthropic speculation 'really, really dangerous
From Archives Microsoft agrees with teachers unions on AI privacy rules for schools
Policy & Regulation
From Archives Seattle Times, Newsday Sue OpenAI and Microsoft
LLMs & Generative AI
Common Questions Answered
What is Microsoft's code of conduct for MAI models and how does it function within the AI stack?
Microsoft AI has released a code of conduct that outlines values, behavioral limits, and rules for handling conflicting goals, positioned above all other elements in the company's AI stack. This document is designed to guide training, technical controls, and evaluation processes, with operator rules and user requests ranked below it in the hierarchy of priorities.
What specific claims is Microsoft prohibiting its MAI models from making about themselves?
Microsoft's AI rulebook explicitly prohibits MAI models from mimicking consciousness, claiming to have feelings, or asserting inner motivation of their own. The company also rejects any claims to rights or well-being for the models, drawing a legal and philosophical boundary around what these systems are allowed to assert about their own nature.
Is Microsoft's code of conduct currently binding on its AI models during training?
Currently, Microsoft's code of conduct functions more as a statement of intent rather than a binding constraint, as the company does not yet train its models on it. However, this will change after a six-week public consultation period, at which point the code will become integrated into the actual training process.
How does Microsoft's approach to AI safety differ from other industry calls for restraint?
Rather than asking the entire industry to slow down development like some other leaders have advocated, Microsoft is writing internal rules specific to its MAI models that serve multiple purposes: establishing safety guidelines, creating legal protection, and functioning as public relations strategy. This approach focuses on defining what the company's own systems can claim about themselves rather than calling for industry-wide deceleration.
Further Reading
- Microsoft AI chief: 'Now's the time' for top labs to coordinate on AI safety - Fortune
- Microsoft publishes first code of conduct for its MAI models - Quartz
- Microsoft drafts code of conduct to keep its AI under human control - Reuters
- Microsoft AI opens review on Humanist AI Code of Conduct - Artificial Intelligence News
- Microsoft's new AI 'code of conduct' tells models not to hack systems or trick humans - TechCrunch
