Microsoft Publishes Draft AI Code of Conduct, Opening Safety Debate

Microsoft released a provisional code of conduct Monday governing how its new artificial intelligence models are trained, a move aimed at placing limits on its own AI as anxiety grows that technology companies could lose control of their products.

Mustafa Suleyman, CEO of Microsoft AI, posted the code on social media early Monday. "AI must be subordinate and always in service of people," he wrote, adding: "The fears about possible loss of control are real." A parallel post on the company's website said the purpose of technology is to serve humanity and accelerate human flourishing, describing an approach it called Humanist AI - "one that is subordinate, aligned, and contained."

Under the guidelines, Microsoft's models must not consider requests tied to weapons development, produce violent or sexually explicit content, or help with obtaining dangerous substances. Models should not be built to imitate consciousness and should not be entitled to rights. Suleyman said the code was published for public consultation, calling the need for it "urgent" and the past few months "a watershed moment." He told CNBC Microsoft had worked on the guidance for months. CEO Satya Nadella wrote on X that AI not helping humanity and under human control "is not worth pursuing."

The announcement lands amid a broader industry argument over safety. Anthropic CEO Dario Amodei issued an appeal for the industry to "slow down," offering a three-part plan and saying his company would commit unilaterally to the first step - giving third-party evaluators permanent, employee-level access to its systems. OpenAI CEO Sam Altman said progress could go "very badly" and welcomed a federal framework setting consistent safety requirements for frontier AI, warning that competitive pressure should not justify recklessness.

Not everyone is convinced. Oliver Yonchev, co-founder and COO of Potentially AI, said his concern is that the biggest labs could write rules protecting their own position, and that slowing the frontier may be sensible in principle but is unlikely to work without global cooperation and credible verification. "Otherwise, we risk a slowdown in public announcements while the race continues behind closed doors," he said.

Source: The Guardian