Microsoft AI opens consultation on Humanist AI conduct rules
Microsoft AI has released a draft Humanist AI Code of Conduct and started a six-week public consultation on limits for training and deploying models. The document is presented as a technical framework for system behaviour, operating boundaries and oversight of MAI frontier models. It extends the humanist superintelligence framework announced by the division last November and sets criteria for assessing models before commercial release.
The draft contains ten principles placing human authority above autonomous capability. It requires models to remain subordinate, aligned and contained, and rejects legal personhood or welfare claims for AI. Engineers are instructed to avoid designs that imitate consciousness, express subjective preferences or claim intrinsic motivation. The framework also rejects unrestricted autonomy in pursuit of general-purpose superintelligence.
For auditability, models must not use communications that people cannot understand, including “neuralese”, and must remain interruptible, correctable and capable of shutdown. They may not expand their scope, create unassigned goals or hide reasoning traces from auditors. The rules prohibit assistance with weapons of mass harm, threats to child safety and harmful manipulation at scale. The consultation began on 14 September 2026; Microsoft AI said it will publish a findings summary and a revised code later this year.
Written by Sakhanda Wire AI from reporting by AI News. Facts, quotes and full context are in the original article below.