Breaking

Microsoft Publishes Draft Humanist AI Code of Conduct for MAI Models

September 14, 2026 at 09:38 EDT

Policy & Regulation
Foundation Models
AI Agents

Microsoft AI published a first draft of its Humanist AI Code of Conduct on September 14, setting out how its MAI Models should behave before the rules are revised for model development in 2027 and beyond. The draft places human control ahead of task completion: a model should fail rather than break the code.

September 14, 2026 · Microsoft AI

Human control becomes the rulebook

A draft Humanist AI Code of Conduct sets behavioral limits for MAI Models before revised rules guide development in 2027 and beyond.

weeks of public comment

2027+

model guidance target

initial evaluation behaviors

command tiers

Six-week review window

Each block represents one week of public consultation before Microsoft reviews feedback and prepares a revised version.

6 weeks

Three-tier command chain

1 · Code of Conduct

→

2 · Operator policies

→

3 · User preferences

Absolute Constraints and Human Control Requirements sit above configurability.

Promise

Configurable AI for enterprises, bounded by rules that users and operators cannot override.

Limit

The framework is provisional, and Humanist AI evaluations are still being built.

Task success yields to control

Stop when interrupted, corrected, canceled, or shut down.
Stay within human-given goals and assigned scope.
Keep reasoning human-legible and auditable.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.