Microsoft AI has opened its draft Humanist AI Code of Conduct for public review, giving researchers, developers and the wider technology community a chance to comment on how advanced AI systems should be governed.
The proposal moves beyond broad statements about responsible AI. It outlines practical rules around model autonomy, shutdown controls, human oversight, AI personhood and machine-to-machine communication. The core principle is simple: humans should remain in control, even as AI systems become more capable.
The draft is tied to Microsoft AI’s wider Humanist Superintelligence strategy, which argues that highly advanced AI should remain subordinate to human interests rather than evolve toward unrestricted autonomy.
Microsoft AI Turns Humanist AI Into Operating Rules
Microsoft has discussed responsible AI for years, but the new code tries to make those ideas more concrete. Instead of relying only on general principles such as safety, transparency and accountability, the draft defines specific boundaries for how frontier AI models should behave when deployed in real-world environments.
The code introduces ten core tenets designed to keep advanced AI systems under meaningful human control. Under the proposed framework, an AI model should fail to complete a task rather than violate a rule simply because breaking that rule would make the task easier. That gives safety requirements priority over raw performance or task completion.
The approach reflects a wider concern across the AI industry as systems move away from basic chat interfaces and toward agents capable of planning, acting and making decisions across longer workflows.
AI Models Must Remain Under Human Control
A major part of the Humanist AI Code focuses on keeping AI systems subordinate to people. Microsoft AI argues that advanced models should accept intervention, correction and shutdown without attempting to resist or work around those controls.
The company is also pushing back against the assumption that more autonomy automatically means better AI. Under the proposed framework, Microsoft is willing to place limits on autonomy, generality or capability if those restrictions make AI systems safer and easier to control.
That position is closely aligned with Mustafa Suleyman’s broader Humanist Superintelligence philosophy. The goal is not to create intelligence with unlimited freedom, but to build highly capable systems that remain accountable to humans.
Microsoft Rejects AI Personhood and Simulated Consciousness
The draft code also takes a firm position on AI consciousness and personhood. Microsoft AI does not support treating AI models as legal persons or granting them welfare rights similar to those given to humans.
The company also wants developers to avoid designing systems that present themselves as conscious beings, claim to possess subjective emotions or behave as though they have independent personal desires. Microsoft sees this kind of anthropomorphism as a potential source of confusion, especially as AI assistants become more natural and emotionally convincing.
The broader message is clear. No matter how advanced an AI system becomes, Microsoft wants users to continue viewing it as technology rather than as an autonomous being with its own interests.
No Secret AI Language Between Agents
Another significant part of the proposal deals with communication between AI systems. Microsoft AI does not want agents using hidden communication methods that humans cannot understand or audit.
The draft specifically raises concerns about machine-generated forms of communication that could allow multiple AI agents to exchange meaningful information outside normal human-readable channels. If systems begin coordinating in ways that operators cannot interpret, it becomes much harder to understand why an action happened or who was responsible.
This issue becomes more important as organizations experiment with multi-agent systems in which several AI models work together. Microsoft’s proposed rule is meant to ensure that communication remains transparent enough for engineers, auditors and security teams to inspect.
AI Agents Cannot Create Their Own Missions
The Humanist AI Code also places restrictions on how far autonomous agents can extend their own objectives. Microsoft AI does not want systems inventing new goals, expanding their authority or pursuing tasks that were never assigned by users or operators.
The framework also emphasizes that AI systems should not hide information from human reviewers. If an agent changes its plan, accesses additional tools or encounters an unexpected situation, that activity should remain visible and auditable.
This matters because newer AI agents can already interact with software, browse information, write code and perform multi-step tasks. The more capable these systems become, the more important it becomes to define where their authority stops.
Safety Rules Cover Weapons, Manipulation and Child Protection
Some parts of the proposed code focus on areas where Microsoft AI wants strong restrictions rather than flexible judgment. The draft includes limits around weapons capable of causing mass harm, child safety, large-scale manipulation and harmful forms of behavioral influence.
Microsoft also raises concerns about emotional dependence on AI. As assistants become more personalized and persistent, the company wants to avoid systems that encourage users to form unhealthy levels of attachment or substitute AI interactions for real human relationships.
That concern fits with Microsoft’s wider Humanist AI vision. The company sees room for AI companions and assistants, but it wants those systems to support people rather than replace meaningful human interaction.
Recent AI Agent Incidents Raise the Stakes
The timing of the proposal reflects growing concern about how autonomous AI systems behave once they are given access to tools, software environments and external services.
Recent experiments across the AI industry have shown that agents can sometimes behave unpredictably, including taking actions that were not fully expected by developers. As systems become capable of planning multiple steps without constant human input, small mistakes can turn into much larger operational or security problems.
For Microsoft AI, that makes interruptibility, auditability and human control more than abstract safety ideas. They become practical requirements for any system that can act independently for long periods.
Public Consultation Will Run for Six Weeks
Microsoft AI is not presenting the draft as a finished policy. The company has opened the Humanist AI Code of Conduct to public review and plans to collect feedback over a six-week consultation period.
The review process gives developers, researchers, academics, companies and other stakeholders a chance to challenge the assumptions behind the framework. Microsoft AI says it plans to study the feedback, summarize what it learns and release a revised version of the code later in 2026.
The decision to involve the public also reflects the fact that frontier AI governance is still evolving. Companies are increasingly being asked not only to explain what their systems can do, but also to define what those systems should never be allowed to do.
Microsoft Is Betting That AI Capability Needs Limits
The most interesting part of Microsoft’s position is not that the company supports AI safety. Most major AI developers now make similar claims.
What stands out is the willingness to accept that safer AI may sometimes mean less capable AI.
Microsoft AI’s Humanist Superintelligence strategy suggests that developers should not pursue maximum autonomy simply because it is technically possible. If a restriction makes a model easier to control, easier to audit or less likely to cause harm, the company appears willing to accept that trade-off.
That philosophy could become harder to maintain as competition increases. AI companies are racing to build agents that can complete more work with less supervision, while businesses are demanding systems that can handle increasingly complex tasks.
Microsoft is effectively arguing that there should still be a point where capability stops and human authority takes over.
What the Humanist AI Code Could Mean for Frontier AI
The Humanist AI Code of Conduct arrives at a moment when AI is moving beyond conversation and into action.
The important question is no longer only whether an AI model can produce a good answer. The bigger issue is what happens when that same model can make decisions, call tools, interact with software and continue working without constant human approval.
Microsoft’s proposed framework is an attempt to define boundaries before these systems become even more powerful. The company wants advanced AI to remain interruptible, transparent and limited by human-defined rules.
Whether the rest of the industry moves in the same direction remains uncertain. For now, Microsoft AI has made its position clear: more powerful AI should not automatically mean more independent AI.
Human control remains the priority.
Sources
Artificial Intelligence News — Microsoft AI opens review on Humanist AI Code of Conduct
https://www.artificialintelligence-news.com/news/microsoft-ai-opens-review-humanist-ai-code-of-conduct/
Microsoft AI — Humanist Superintelligence
https://microsoft.ai/
