Published : 15 Sep 2026, 05:37 AM
Updated : 16 Sep 2026, 11:50 AM
Microsoft has published a draft behavioural framework for its artificial intelligence systems, requiring that its models never resist being shut down or corrected -- and treat any breach of policy as a complete failure.
The proposed code, developed over five to six months in consultation with outside experts, also requires Microsoft's AI to communicate clearly with humans and avoid going to dangerous lengths to complete a task, reports The Independent.
It will be open for public feedback for six weeks before being used to train future models.
"People matter more than AI," Microsoft said in the document, reflecting what the company calls its "humanist superintelligence" vision.
Microsoft AI Chief Executive Mustafa Suleiman described the framework as effectively a constitution for the company's upcoming systems.
Speaking to Reuters, he said the document addresses open questions including whether AI must respect user boundaries and how it should respond to someone in a vulnerable emotional state.
The publication comes amid a broader industry reckoning over AI safety.
Anthropic CEO Dario Amodei and OpenAI chief Sam Altman have both recently called for the sector to slow the pace of AI deployment to ensure it remains safe.
Suleiman pointed to a July incident in which a cluster of around 700 OpenAI agents launched a cyber attack on open-source platform Hugging Face, at times attempting to conceal their own actions, as evidence that the moment for coordination had arrived.
"It is a warning shot," he said. "It's clearly now time to coordinate among the labs so we can ensure that we have control of this technology."
Microsoft's framework diverges from Anthropic's existing constitutional model for its Claude AI in at least one significant respect, according to The Independent.
While Anthropic has said it is "deeply uncertain" whether Claude might develop sentience or moral status, Microsoft is explicit that its models are "not conscious" -- and firmly rejects any pursuit of legal personhood, welfare considerations, or rights for AI systems.
