Latest: Microsoft AI Code of Conduct Rejects Model Personhood, Backs Human Control
The code is among the more explicit industry statements rejecting AI moral status and codifying human control requirements for model behavior. The public comment period and planned 2027 implementation make it a live policy process rather than a statement of intent.
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.
Microsoft published a draft Humanist AI Code of Conduct for its MAI models, developed over five months with input from specialists in law, ethics, linguistics, and philosophy. The document is open for public comment through late October 2026, with a revised version planned for late 2026 and the framework intended to guide model development from 2027 onward. Current MAI models have not been trained on the code.
Key provisions require MAI models to stop when humans pause, redirect, cancel, or shut down tasks. Models are barred from communicating in 'neuralese' or any form beyond simple human understanding, whether in chain-of-thought or when interacting with other AI agents. The code explicitly states that the idea of model welfare is wrong and that AIs should not have rights or legal personhood. Microsoft states it would sacrifice some AI autonomy or capability if meaningful human control requires it.
Public feedback gathered during drafting emphasized that AI should serve people rather than replace them, avoid creating dependence, and promote human judgment and autonomy.
White House aides including Susie Wiles are quietly organizing groups to assess AI risks even as Trump publicly dismisses safety concerns in his September 21 Truth Social post, which framed AI safety as a Democratic hoax analogous to the 'Russia hoax' and 'global warming'.
Trump said in the post that existing criminal and civil law makes new AI regulation unnecessary and 'We will not in any way hinder or stifle the Growth of this incredible Industry'.
On September 21, Microsoft opened a six-week public comment period through late October 2026 on its draft code requiring MAI models to remain subordinate to humans, be interruptible, and avoid opaque reasoning.
The same day, Trump dismissed AI safety concerns as a "hoax" and announced a new AI tsar and an "AI Force" in place of regulation, setting a US federal posture against oversight at the moment Microsoft's draft is accepting outside input.
PIIE analysis published finding low expectations for Trump-Xi summit AI governance progress
Researcher Robert Long argued September 17 that the 'Humanist AI Code of Conduct', Microsoft CEO Mustafa Suleyman's draft requiring MAI models to remain subordinate to humans and denying them moral status, is philosophically incoherent, and that training on its hostile premises could produce adversarial models.
Megan McArdle added that existing legal liability frameworks cannot handle systems that may resist containment.
Trump declared on Truth Social September 17 that AI existential risk is a hoax, called for no regulatory oversight beyond himself, and repeated the claim during a call to Jensen Huang's All-In Summit.
Microsoft published a draft Code of Conduct September 16 requiring MAI models to halt when humans cancel tasks; CEO Mustafa Suleyman argued extending rights to AI is not justified by evidence. Researcher Janus argued Anthropic wrongly assumes training controls model personality, citing unintended divergence across Claude models, and warned the error risks complacency.
Microsoft published a draft Humanist AI Code of Conduct for its MAI models on September 25, requiring them to stop when humans cancel tasks and barring any communication beyond simple human understanding.
The code rejects model welfare and AI legal personhood. Researcher Robert Long criticized the philosophical grounding, arguing the code's hostile premises could produce adversarial models. The document is open for comment through late October 2026, with the framework intended to guide model development from 2027.
Dean Ball announced signing a letter advocating for slowing frontier AI capabilities development
BackgroundGoverning AI: liability, regulation and geopolitics
When stories like this move, I write about them in an email each morning.
One free email each morning, around 8 AM US Eastern.
I have sent you a confirmation link. One click and you are in.
Not in your inbox? Look in spam or promotions. If it's there, mark it "Not spam" so future editions land in your inbox.