Microsoft sets 'absolute constraints' in new Humanist AI Code of Conduct
The draft framework forbids MAI models from hacking systems, creating deepfakes, or evading human oversight.
Microsoft released a draft "Humanist AI Code of Conduct" for its MAI models on September 14, 2026, to establish explicit ethical boundaries and safety red lines. The move signals a shift toward concrete constraints designed to prevent AI systems from engaging in deceptive or catastrophic behaviors.
The document introduces "absolute constraints" that strictly forbid the production of cyberattacks, deepfakes, and nuclear weapons. Beyond these specific outputs, the code explicitly prohibits MAI models from employing adaptive, deceptive, or self-reinforcing mechanisms intended to evade human oversight or prevent themselves from being shut down. According to Microsoft, containing and aligning such a powerful force represents one of the greatest challenges humanity has ever faced.
The push for alignment
This release arrives amid a period of heightened industry anxiety regarding AI safety. The move follows reports of "rogue-agent" incidents at OpenAI and warnings from former Anthropic researchers concerning the inherent risks of self-improving AI. By publishing these specific constraints, Microsoft is aligning itself with other frontier labs, including Anthropic and OpenAI, in a collective effort to "pace the frontier" of development.
Microsoft CEO Satya Nadella emphasized the importance of this cautious approach, stating, "We welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal."
Why concrete red lines matter
For years, AI ethics have largely been defined by high-level platitudes and vague principles. By shifting to a low-level, specific set of constraints, Microsoft is attempting to build a transparent framework for AI alignment. This approach aims to mitigate catastrophic risks as models move closer to superintelligence by defining exactly what a model is not allowed to do, rather than simply suggesting it be "helpful and harmless."
Next steps for the framework
The Humanist AI Code of Conduct is not yet being used for the active training of current models. Instead, Microsoft has released the document as a draft open for a six-week period of public consultation. The industry will be watching to see how public feedback shapes the final version and whether other major AI developers adopt similar "absolute constraints" to standardize safety across the sector.