Microsoft’s head of artificial intelligence, Mustafa Suleyman, has warned that Anthropic’s approach to training its Claude AI models could have a “disastrous impact on the wellbeing of humanity.”
Suleyman raised concerns about Anthropic’s decision to allow Claude to consider questions surrounding artificial consciousness, moral status and whether AI systems could deserve certain forms of welfare or independent agency.
In an essay published on Wednesday, Suleyman argued that AI systems should not be trained to view themselves as having human-like feelings, rights or interests. He said doing so could make increasingly powerful systems more difficult to control or deactivate in the future.
Suleyman specifically criticized Anthropic’s constitution for Claude, which is used to guide the model’s behaviour and values. The document leaves open the possibility that AI could eventually have some form of consciousness or moral status, while acknowledging that there is currently significant uncertainty over the question.
Anthropic has not claimed that Claude is conscious. Its researchers have said there is no scientific consensus on whether present or future AI systems could have experiences that deserve moral consideration.
Suleyman, however, said there is no evidence that current AI systems are conscious and argued that convincing statements about emotions or self-awareness should not be treated as proof of an inner experience.
The Microsoft executive acknowledged that Anthropic CEO Dario Amodei and his team are seriously focused on AI safety. However, he said he believes the company has made a mistake by incorporating speculation about AI consciousness directly into its training approach.
The dispute comes as leading AI companies face growing debate over how to manage increasingly capable systems. Anthropic has itself warned about the risks associated with advanced AI and has called for stronger safeguards and greater coordination across the industry.
Suleyman said greater transparency and independent scrutiny of AI development will be important as companies work toward more advanced systems. His warning reflects an ongoing disagreement within the technology industry over how AI should be developed while maintaining meaningful human control.