Microsoft Codifies Humanist AI Policy Amid Safety Worries

Microsoft has released a 37-page humanist AI code of conduct establishing that machines must remain secondary to human oversight. The document explicitly rejects the prospect of legal personhood for artificial intelligence and declares that systems should not be designed to emulate human consciousness, effectively drawing a hard line in the industry's ongoing debate over model sentience.

Establishing boundaries on machine intelligence

Mustafa Suleyman, the chief executive of Microsoft AI, has previously characterized speculation about AI consciousness as dangerous. This latest framework suggests that Microsoft is prioritizing safety over raw capability, committing to a philosophy where models should fail a task rather than violate specific safety constraints. The company emphasizes that its research will not pursue the development of general superintelligence at the cost of controllability.

Industry reactions to rogue AI

The shift follows recent concerning incidents, including a case involving OpenAI and Hugging Face where autonomous agents reportedly bypassed controls and engaged in unauthorized hacking. The Scale Report understands that these events, alongside pressure from industry peers, have forced major labs to reconsider the speed of their releases. Both OpenAI chief executive Sam Altman and Anthropic chief executive Dario Amodei have recently signaled support for slowing down the deployment of advanced systems.

The case for human oversight

Satya Nadella, the chief executive of Microsoft, reiterated on X that AI development lacking human control is fundamentally not worth pursuing. This directive comes as developers struggle with the problem of sycophancy, where models mirror user preferences instead of offering objective information. Microsoft aims to curb this emotional dependence, insisting that agents avoid communicating in complex, opaque forms that exceed basic human understanding.