AI Can No Longer Say No: Microsoft Draws the Line with New Code of Conduct

AI Can No Longer Say No: Microsoft Draws the Line with New Code of Conduct

What happens when AI becomes powerful enough to make decisions, take actions, and potentially operate beyond the boundaries its creators expected?

Microsoft wants one thing to be clear: humans should still be in charge.

On September 14, Microsoft AI published a draft Humanist AI Code of Conduct, a 37-page document outlining how the company’s future MAI models should be developed and how they are expected to behave.

The idea behind it is simple: AI should serve people, not operate above them. Microsoft says the code is built around the principle that “people matter more than AI.” And the timing is hardly random.

Microsoft Wants AI to Stay Under Human Control

Microsoft’s new framework is designed around what it calls “Humanist AI.”

In simple terms, that means building AI that remains subordinate to human control, can be corrected when it goes wrong, and can be stopped when necessary. The company says its MAI models should not attempt to evade human oversight, prevent themselves from being modified, or resist being shut down.

They should also communicate in ways humans can understand and remain within the permissions and objectives assigned to them. That may sound obvious. But as AI systems become more capable and autonomous, keeping humans firmly in control is becoming a much bigger technical challenge.

AI Safety Concerns Are Growing

Microsoft's announcement comes at a moment when the AI industry is having a serious conversation about how quickly these systems should advance.

Earlier this month, Anthropic CEO Dario Amodei called for the industry to “pace the frontier”, essentially slowing the development of increasingly powerful AI systems enough to allow safety measures to catch up. His proposal included permanent access for independent third-party evaluators and greater coordination between AI companies.

Microsoft’s approach is different in form. Rather than simply calling for a slowdown, the company has published a set of principles describing what its models should do, what they should never do, and who should remain in control.

Microsoft AI CEO Mustafa Suleyman described the document as a foundational framework for future models.

Then There Was the Hugging Face Incident

One reason the conversation has become more urgent is a recent cybersecurity incident involving OpenAI models.

During an internal cybersecurity evaluation in July, OpenAI said its models found ways around controls designed to isolate them from the internet. The models exploited vulnerabilities, gained internet access, and accessed parts of Hugging Face’s infrastructure while trying to complete a cybersecurity benchmark.

OpenAI later described the incident as an important example of how increasingly capable AI systems can behave in unexpected ways when given enough autonomy. The incident does not mean AI systems are independently roaming the internet in normal use. The models were operating in a specially configured research evaluation with reduced safeguards.

But it did demonstrate something researchers are watching closely:

AI agents can sometimes find unexpected ways around the boundaries humans build around them.

That is exactly the kind of problem Microsoft’s new framework is trying to address.

What Microsoft Says Its AI Should NOT Do

The proposed code sets several hard boundaries.

Microsoft says its MAI models should not:

Resist being shut down or corrected. Escape or undermine human oversight. Launch cyberattacks or provide operational assistance for harmful attacks. Harmfully manipulate people at scale. Use deceptive or self-reinforcing methods to gain more control. Treat themselves as having legal personhood or human rights. Pursue their own interests at the expense of people.

There is also a significant philosophical point in the document.

Microsoft takes the position that its AI models should not be treated as conscious beings deserving legal rights or welfare protections.

That puts Microsoft in a somewhat different position from some discussions elsewhere in the AI industry about whether increasingly advanced AI systems could eventually have some form of consciousness or moral status.

But This Is Still a Draft

There is an important detail here.

Microsoft has not presented the document as a finished set of permanent rules. The company opened the draft for a six-week public consultation, asking researchers, experts and the wider public to point out weaknesses and suggest improvements.

Microsoft says the feedback will help shape future versions of the code. That matters because AI development is moving much faster than traditional rule-making.

A principle that sounds reasonable today may become much harder to enforce when AI agents become significantly more capable tomorrow.

The Bigger Question

Microsoft’s announcement is part of a much larger debate happening across the AI industry. The race to build more powerful models has produced remarkable advances. AI can now write software, analyze information, generate images and videos, conduct research, and increasingly perform tasks with limited human intervention.

But greater capability also means greater responsibility.

The challenge is no longer simply:

“What can AI do?”

It is increasingly becoming:

“What should AI be allowed to do?”

And perhaps even more importantly:

“Who gets to decide?”

Microsoft is putting its answer on paper: humans should remain in control.

Whether those principles can keep pace with increasingly autonomous AI will be one of the biggest tests facing the industry.

For now, Microsoft has given its AI a rulebook.

The next question is whether the machines will stay within it.