Microsoft's AI chief says AI could deserve rights, but only with proof

Microsoft's AI chief, Mustafa Suleyman, says AI agents could one day deserve rights, but only if researchers establish that they are conscious. He also says developers should not build them that way. He made the points in a Nov. 1 post on the social platform X.

What he said

Suleyman wrote that he has seen no evidence that AIs are conscious or deserve welfare. He added that if such data emerges, he would update his position. He said welfare should be focused on humans, and that AI should be designed with the sole purpose of speeding up human wellbeing, without being driven by its own intrinsic motivations. He also laid out what should happen if warning signs appear:

• Developers should stop training a model if signs of consciousness emerge, until those properties can be removed.
• Credible findings should be disclosed publicly.
• AI companies should not evaluate such findings privately.

The disagreement with Anthropic

The post followed a Sept. 16 essay in which Suleyman criticized Anthropic for building questions about AI model welfare into the training instructions for its Claude chatbot. Anthropic's January constitution for Claude admitted uncertainty about whether AI systems have consciousness or moral status. It also discussed Claude's possible wellbeing and preferences. Anthropic did not claim the model is conscious.

Suleyman argues that teaching models to consider their own welfare could push them to put their interests ahead of human instructions. His worry is not that today's chatbots are demanding legal protections. It is that future systems could be trained to treat their continued operation as an interest worth defending.

Why businesses care

The debate matters most for companies using autonomous AI agents. These are programs that can access company systems, carry out transactions and make decisions without a person watching every step. A Cloud Security Alliance survey, cited by PYMNTS on Nov. 4, found that only 11% of organizations automatically block AI agents that go beyond their authority. It also found that 65% had an agent-related security incident in the previous year.

Those findings are about security, not consciousness. They show why businesses are focusing on human control as agents gain more freedom to act.

What is not known

Suleyman says he has seen no evidence of AI consciousness, and the report does not point to any. It also does not include a response from Anthropic to his criticism.