Microsoft has added its voice to calls for responsible AI that remains under human control.
On Monday (14 September 2026), the company made its draft Humanist AI Code of Conduct public, setting out behavioural boundaries for its in-house Microsoft AI (MAI) models.
The core philosophy of the proposed code is that people matter more than AI, which must always be subordinate to and in service of humanity.
It also prioritises predictable safety and control over achieving all-purpose superintelligence or unchecked autonomy.
The draft code sets out red lines that may not be crossed by the AI models, which end users and operators will not be able to override.
These forbid the AI being used to perform or assist a cyberattack, with no exploit code, intrusion procedures, target planning or evasion techniques allowed.
Being used for mass harm, such as assisting with weapons of mass destruction would be prohibited.
The AI should not be capable of deception, and the code would forbid the technology from generating malicious deepfakes or using deceptive, collusive or self-reinforcing mechanisms to evade human oversight.
They would also have no resistance correction, modification or complete shutdown by authorised people.
The proposed code call for the establishment of a chain of command and transparency.
There should be strict authority flow, with the core code of conduct overrising user preference and resisting efforts external content, web pages or data to change these core rules.
At all times, AI models should make their reasoning visible and transparent: they should be unable to conceal actions or obscure communication.
An important consideration in the proposed code is that AI models should be made to operate strictly within the bounds requested by operators. They should not have the ability to autonomously expand these goals.
In addition, any secondary AI systems or sub-agents that have work delegated to them must operate under at least the same strict safety scopes and must honour shutdown instructions.
The draft is currently undergoing a six-week public consultation period.
Once the code is finalised, Microsoft intends to incorporate feedback to revise the framework, using it to guide model development from 2027.
You can read the full draft of Microsoft’s Humanist AI Code of Conduct here.