Microsoft has published an interim behavior code for its next generation of artificial intelligence models, putting a formal set of rules around how the systems should respond and what kinds of autonomy they should avoid. The company says the work began about five months ago with input from specialists in law, ethics, linguistics and philosophy.
A central principle is that models should not create or pursue independent goals of their own. The code also sets restrictions on assistance involving weapons, dangerous materials and other harmful requests. Microsoft plans to gather external feedback before issuing a final version intended to guide model development from 2027, making the current document a draft governance instrument rather than a finished technical standard.
The timing matters because AI companies are under growing pressure to show that safety rules can keep pace with more capable systems. A written code does not prove that a model will follow it reliably, so implementation and testing will matter more than the language alone. The useful next evidence will be how Microsoft converts the principles into training, evaluations and product-level enforcement.