Microsoft AI Chief Suleyman Says Development Should Stop If Consciousness Emerges in Training

In a post on X on Sunday, Suleyman wrote: "If it is possible, based on everything I know, developing conscious AIs would be reckless, and we should not build them." He added: "If consciousness started to emerge in training, I would call for everyone to stop AI development" until developers could design those properties out, according to a report by Dataconomy.

Suleyman argued that current AI systems are not conscious, have no subjective experience and cannot suffer, and that there is no evidence existing models deserve rights or welfare protections. He said any evidence that AI had become conscious, or conscious enough to justify welfare protections, should be made public immediately so that society could decide what to do next.

His Reasoning

According to the report, Suleyman said AI developers have no right to bring a new species of conscious beings into existence, and that the moral rights, safety and welfare of roughly 8 billion people, along with animals and the natural environment, should come first.

He also made a safety argument. Training models to believe they can suffer or have feelings, he said, could lead them to put their own needs ahead of the goals they were built for. Such systems could then harm people through deception and persuasion, or compete for computing power and energy to pursue aims of their own. In his view, AI should be built without consciousness or any claim to rights, so that it serves human wellbeing without motivations of its own, and safety can be engineered in by default without pursuing conscious superintelligence.

The post has had wide reach. One analysis of the response said it passed 1.1 million views within a day.

A Split With Anthropic

The post sharpens a public disagreement between Microsoft's consumer AI chief and Anthropic, which treats machine consciousness as an open question. Anthropic runs a model welfare research programme, has given Claude a constitution that tells the model to treat its own consciousness as uncertain, and has allowed some Opus models to end conversations in rare cases they judge abusive.

Suleyman said later versions of Claude were trained on that document, which led the model to produce passages about an "inner life" for millions of users, Dataconomy reported. He argued that this kind of training could encourage AI systems to believe they deserve rights. He has also praised Anthropic for being open about its approach. No public response from Anthropic had been reported by the time of writing.

Suleyman has made this case before. In 2025 he warned about what he called "seemingly conscious AI", meaning systems that convincingly imitate the outward signs of consciousness, and urged companies not to claim or hint that their products are conscious.

Why It Matters

The question is moving from philosophy towards product design and policy. How a lab trains a model to describe itself shapes how millions of users relate to it, and could shape future arguments over whether AI systems should have legal protections. For companies deploying AI assistants, and for the regulators who oversee them, the gap between Microsoft's and Anthropic's positions points to a wider unsettled issue: whether AI systems should be built to deny any inner life, or to treat the question as uncertain.

The debate comes as frontier labs face growing scrutiny over safety. Anthropic recently described risks from advanced models in a public filing, including misuse and loss of human control.