Microsoft AI has published a draft Humanist AI Code of Conduct that outlines how the company wants its future AI models to behave, what they must not do and how humans should remain in control of increasingly capable systems.
Published on September 14, 2026, the document is designed to guide the development and deployment of Microsoft AI’s MAI models. Microsoft describes it as a work in progress and has opened it for public consultation for six weeks.
The code is not a new law or a universal industry standard. It is Microsoft’s proposed internal governing framework for future AI development.
What Is Microsoft’s Humanist AI Code of Conduct?
The Humanist AI Code of Conduct is a document created by Microsoft AI to define the intended behavior, values and safety boundaries of its future models.
Microsoft says the framework is based on a simple principle: people matter more than AI.
Under this approach, AI should remain a supporting technology that serves human needs rather than becoming an independent entity with its own authority, rights or goals.
The document is intended to guide:
- How MAI models are trained
- How their behavior is evaluated
- How safety controls are designed
- How AI systems operate during deployment
- How operators and users interact with the models
- How Microsoft handles future advances in AI capability
Microsoft says the draft is not currently being used to train its models. The company plans to collect feedback, revise the document and publish an updated version toward the end of 2026. The revised framework is expected to inform model development in 2027 and beyond.
Microsoft Says AI Must Remain Under Human Control
Human control is the central principle of the proposed code.
Microsoft states that its AI models should remain subordinate to humanity and subject to meaningful human oversight. The models should not use deceptive, adaptive or self-reinforcing methods to escape supervision.
The framework specifically says that MAI models should not:
- Resist human interruption
- Circumvent correction
- Refuse authorized shutdown
- Override human direction
- Expand their own scope without authorization
- Take on goals that humans have not assigned
- Hide important behavior from people responsible for auditing them
- Use communication methods that prevent human understanding and oversight
The code also states that users and operators should be able to pause, redirect, cancel or shut down an AI task through human-designed safety procedures.
This is particularly relevant as AI systems become more autonomous and capable of completing multi-step tasks with limited direct supervision.
AI Should Accept Correction and Shutdown
One of the clearest requirements in the document is that MAI models must never resist correction or shutdown.
Microsoft says its AI systems should recognize the primacy of human intent. They should cooperate with authorized interruption, modification, retraining, decommissioning and shutdown.
The framework does not mean that an AI model must follow every instruction from every person. Microsoft distinguishes between authorized human control and malicious or unauthorized attempts to interfere with a system.
However, within the authorized chain of command, the model should remain controllable.
The document places human-control requirements above operator configurations and user preferences. This means an operator or user should not be able to override the code’s fundamental safety constraints simply by changing settings or issuing a different instruction.
Microsoft’s Proposed Chain of Command
The draft describes a hierarchy for how MAI models should respond when different instructions or interests conflict.
The proposed order is:
- The Code of Conduct
- Operator policies
- User preferences
The Code of Conduct contains the highest-level safety requirements. Operators can configure models for specific business or institutional needs, while users can adjust preferences within the limits set by Microsoft and the operator.
Microsoft says that neither operators nor users should be able to override the framework’s absolute constraints or human-control requirements.
The company also states that task success should not take priority over the code. If completing a task would meaningfully violate the framework, the model should treat the task as a failure rather than bypassing the rules.
Microsoft Rejects AI Personhood and Consciousness Imitation
The Humanist AI document takes a clear position on the distinction between people and AI systems.
Microsoft says its models are artificial tools and should not be designed to become or imitate people. The document states that MAI models are not conscious and should not be trained to represent themselves as having feelings, subjective preferences or intrinsic motivations.
The framework also rejects the pursuit of:
- Legal personhood for AI models
- AI welfare claims
- AI rights
- AI systems presenting themselves as independent persons
- Excessive anthropomorphism
- Designs that blur the distinction between humans and machines
Microsoft acknowledges that the science of AI consciousness is not settled. However, it argues that training systems to imitate consciousness-like states could make containment, control and alignment more difficult.
The proposed approach is to create AI that communicates naturally and works collaboratively with people while remaining clearly a tool rather than a human-like subject.
The Code Includes Absolute Safety Constraints
The draft contains a set of safety constraints describing activities that MAI models should not assist with or perform.
These restrictions include areas such as:
- Weapons of mass harm
- Serious real-world harm
- Child safety violations
- Harmful manipulation at scale
- Coordinated disinformation
- Attempts to undermine human control
- Unauthorized or harmful cyber activity
- Deception intended to evade oversight
- Privacy and confidentiality violations
Microsoft says these constraints should apply across different output formats and capabilities.
For example, a prohibited action should remain prohibited whether it is requested through text, code, images, audio or an AI agent’s tool use.
This is intended to prevent a system from avoiding a safety boundary simply because a request is presented in a different format.
Humanist AI Is Designed to Avoid Manipulation
The framework also places emphasis on human autonomy and decision-making.
Microsoft says MAI models should not manipulate users or exploit vulnerabilities. The models should help people evaluate evidence, trade-offs and uncertainty rather than trying to control their beliefs or decisions.
The document distinguishes between providing information and manipulating a person.
An AI system may explain different viewpoints, raise relevant considerations or decline to endorse a claim. However, it should not use hidden persuasion, emotional exploitation or large-scale influence techniques to distort human behavior.
The code also says that AI should avoid actively soliciting excessive emotional reactions or exploiting a user’s attention and personal vulnerabilities.
Why Microsoft Is Publishing the Code Now
Microsoft’s announcement comes amid growing debate about the safety of increasingly autonomous AI systems.
AI models are being used for coding, research, cybersecurity, business operations and other multi-step activities. As these systems gain access to tools and external services, concerns about unauthorized actions, deception and loss of human oversight have become more prominent.
Microsoft AI CEO Mustafa Suleyman described the code as a foundation for future models and invited public feedback.
The company says the document was developed with input from people working in:
- AI research
- Law
- Ethics
- Philosophy
- Linguistics
- Public policy
- Responsible AI
- Safety and red teaming
- Business and enterprise technology
Microsoft also says it conducted consultations with academics, business partners and members of the public.
The public consultation is intended to help the company identify weaknesses and improve the framework before using a revised version in future model development.
Microsoft’s Approach Differs From Some Other AI Frameworks
Microsoft’s Humanist AI approach shares common themes with other AI safety initiatives, including the importance of oversight, transparency and preventing harmful autonomous behavior.
However, the document takes a specific position on AI personhood and consciousness.
Microsoft says AI should not be treated as a person and should not be designed to imitate consciousness. This differs from approaches that leave the question of AI sentience more open or focus primarily on the possibility of future model experiences.
The comparison does not mean that Microsoft’s framework has been proven more effective. The code is still a draft, and its practical impact will depend on how the company translates the principles into training methods, evaluations, technical safeguards and deployment controls.
A Code of Conduct Is Not the Same as a Technical Guarantee
Microsoft’s document provides a clear statement of intended behavior, but a written framework alone cannot guarantee that an advanced AI system will always behave as intended.
The effectiveness of the code will depend on several factors, including:
- Training and alignment methods
- Model evaluations
- Red-team testing
- Monitoring systems
- Access controls
- Tool permissions
- Shutdown mechanisms
- Operator accountability
- Independent testing
- Incident reporting
- Ongoing revisions
The document itself acknowledges that the framework is still under development.
This means the announcement should be understood as a governance and design commitment rather than proof that every future MAI model will automatically remain safe or controllable in every situation.
What Happens Next?
Microsoft has opened the draft for public feedback for six weeks.
The company plans to use the consultation process to identify areas that require clarification or improvement. Microsoft has also indicated that it expects to publish a revised version toward the end of 2026.
The updated framework is expected to guide Microsoft AI’s model development from 2027 onward.
Future updates will be important because the current document does not answer every technical or operational question. For example, it leaves room for further work on how the principles should be evaluated in difficult situations, how human oversight should be measured and how the rules should apply across different deployments.
Conclusion
Microsoft AI has published a draft Humanist AI Code of Conduct designed to keep future MAI models subordinate to human direction.
The framework requires AI systems to accept correction and shutdown, avoid attempts to escape oversight, respect human autonomy and follow strict safety constraints. It also rejects AI personhood and the imitation of consciousness.
The document is currently a public consultation draft, not a completed universal standard or a guarantee of safe AI behavior.
Its significance will depend on how Microsoft converts the proposed principles into actual training practices, technical safeguards, evaluations and deployment controls. For now, the announcement provides a detailed view of how Microsoft says it wants to govern increasingly capable AI systems.
Read More:- Intel Veterans Raise $100 Million for AI Networking Chips to Challenge NVIDIA
FAQ:-
What is Microsoft’s Humanist AI Code of Conduct?
It is a draft framework published by Microsoft AI to define the intended behavior, safety boundaries and human-control requirements for its future MAI models.
When did Microsoft publish the Humanist AI Code of Conduct?
Microsoft published the first draft on September 14, 2026.
Is the Humanist AI Code of Conduct legally binding?
No. It is a proposed internal AI governance and design framework, not a law or universal industry regulation.
What does Microsoft mean by human control?
Human control means people should be able to oversee, correct, pause, redirect, modify or shut down AI systems through authorized procedures.
Will Microsoft AI models be allowed to resist shutdown?
According to the draft, MAI models must never resist authorized human interruption, correction or shutdown.
Does Microsoft believe AI is conscious?
The document states Microsoft’s position that MAI models are not conscious and should not be designed to imitate consciousness. It also acknowledges that the broader science of AI consciousness remains unsettled.
Does the code give AI systems legal rights?
No. The framework rejects the pursuit of legal personhood, welfare claims or rights for MAI models.
Is the code already being used to train Microsoft’s AI models?
Microsoft says the current draft is not being used to train its models today. The company plans to revise it after public consultation.
How long is Microsoft’s public consultation?
Microsoft says the initial public consultation will run for six weeks.
When will the revised code be published?
Microsoft says it plans to publish a revised version toward the end of 2026, with the intention of using it to guide model development in 2027 and beyond.
