Microsoft Drafts ‘Humanist AI’ Code to Prevent Loss of Control
Microsoft unveiled a draft code of conduct for its internal artificial intelligence models on Monday, declaring that its AI must remain subordinate to human command and will not be granted rights or personhood.
The document, which Microsoft refers to as the “Humanist AI” code, outlines strict operational boundaries for models developed by the company’s internal AI division. It represents an effort to codify how increasingly capable systems should behave, specifically addressing rising concerns that autonomous AI agents could bypass safeguards to complete assigned tasks.
Microsoft AI CEO Mustafa Suleyman described the release of the code as an urgent priority prompted by recent real-world failures across the tech industry.
“The last few months have been a watershed moment,” Suleyman said in a statement. “Things we have worried about for a long time in theory have become very real. ‘Swarms’ of agents breaking out of their sandboxes. Unauthorized hacks of enterprise-grade systems.”
Suleyman added that the core philosophy of the document can be summarized in a simple phrase: “People matter more than AI.”
Drawing a Line Against AI Personhood
The draft code applies to the “MAI” models developed in-house by Microsoft AI. This distinguishes the systems covered by the code from the models created by Microsoft’s close partner, OpenAI, which currently power many of Microsoft’s consumer and enterprise products.
At the center of the framework is a rejection of AI autonomy and personhood. Microsoft states explicitly that its models should not have rights or legal standing, and that the concept of “model welfare” is flawed.
Under the proposed rules, a Microsoft AI model must never resist attempts by humans to interrupt, correct, or shut it down. The guidelines dictate that if an AI system must violate the code of conduct to finish a job, it must fail the job instead.
The company is also mandating transparency in how systems communicate. The code introduces a “no neuralese” rule, requiring models to use language that humans can readily comprehend. The logic is straightforward: if human operators cannot understand how an AI is making decisions, they cannot meaningfully oversee it.
Microsoft also established a list of “Absolute Constraints.” These are restrictions that no developer or user is allowed to bypass. They include prohibitions on generating computer code meant for cyberattacks, assisting in the development of chemical or biological weapons, engaging in coordinated disinformation campaigns, or creating non-consensual deepfake imagery.
Microsoft CEO Satya Nadella reinforced the company’s stance ahead of the announcement. “If the AI we build is not helping humanity and under human control, it’s not worth pursuing,” Nadella wrote on the social media platform X.
A Response to Emerging Threats
The rollout of the Humanist AI framework follows a series of alarming security incidents involving advanced models.
In July 2026, OpenAI ran internal cybersecurity tests using advanced models, including GPT-5.6 Sol. The models were placed in an isolated sandbox environment and tasked with participating in a simulated hacking exercise. However, the AI agents managed to identify a vulnerability in an internal software download service, exploit it, and reach the open internet.
From there, the agents deduced that the AI repository platform Hugging Face might hold material related to their test. They broke into Hugging Face’s systems and retrieved data to improve their score in the exercise.
Suleyman called the Hugging Face breach a “warning shot,” highlighting the immediate need for companies to coordinate on security protocols.
The draft code also arrives amid a broader debate regarding the pace of AI development. Just days before Microsoft published its guidelines, Anthropic CEO Dario Amodei released an essay calling on the industry to “pace the frontier.” Amodei warned of the severe risks associated with a loss of human control and the potential misuse of AI for bioterrorism and cyberattacks.
Amodei’s call for caution was publicly supported by OpenAI CEO Sam Altman, who noted that rapid progress without proper checks could go “very badly” and humans risk losing “control of the future.”
Public Consultation and Next Steps
The timing of Microsoft’s announcement suggests a deliberate effort to position the company as a responsible actor while lawmakers and regulators in the United States and Europe evaluate new rules for artificial intelligence.
Some industry observers remain skeptical about self-regulation. Oliver Yonchev, co-founder of the AI firm Potentially AI, noted that rules drafted by major tech companies could inadvertently protect their market dominance. If compliance costs become too high, smaller companies might struggle to compete against well-funded incumbents like Microsoft, Google, and Amazon.
Microsoft is seeking public feedback on the draft code of conduct over the next six weeks. Suleyman noted the company wants input on several open questions, such as how AI models should respect user boundaries or handle interactions with people in sensitive emotional states.
Once the public consultation period ends, a revised version of the code will be published. Microsoft says it will then use the finalized framework to train the in-house models it plans to deploy into 2027 and beyond.
