Microsoft AI chief Mustafa Suleyman has reaffirmed his position that today's artificial intelligence systems are not conscious, do not experience feelings or suffering, and do not deserve welfare protections or rights.
However, he has made an important qualification that changes the meaning of his argument: he would reconsider his position if credible evidence emerged showing that AI systems could genuinely experience consciousness or suffering.
In an article published on X on November 1, Suleyman addressed reactions to his earlier comments about AI welfare, arguing that many people had misunderstood what he meant. Rather than claiming machines could never become conscious, he said he was describing the technology as it exists today.
The distinction comes as technology companies face growing questions about whether increasingly sophisticated chatbots could eventually develop experiences that deserve moral consideration.
Suleyman believes developers should avoid deliberately creating conscious machines, but his willingness to revise his position leaves an important question open: what evidence would be enough to change how the industry treats artificial intelligence?
Mustafa Suleyman Explains What Would Make Him Reconsider AI Rights
In his November 1 article, titled 'Don't build conscious superintelligence', Suleyman explained the evidence behind his position.
'I have seen no evidence that AIs are conscious or deserve welfare. If that data emerges, I would of course update my position,' he wrote.
He argued that any credible indication of machine consciousness should be disclosed publicly rather than kept within the companies developing the technology.
According to Suleyman, decisions about the treatment of potentially conscious machines should involve society as a whole, rather than being left exclusively to technology executives and research teams.
He also said that if signs of consciousness began appearing during the training of an AI system, he would call for development to stop until researchers could remove those properties.
Suleyman believes that creating a conscious digital species would raise ethical questions that developers cannot answer on behalf of everyone else.
He described the possibility as an experiment that humanity has not collectively agreed to undertake and argued that protecting existing human lives, animals and the natural environment should remain the priority.
Despite disagreeing with Anthropic's approach to AI welfare, Suleyman praised the company for publishing its constitution and research, saying he respected its technical expertise and commitment to public debate.
The complete statement is available in his article on Mustafa Suleyman's X account.
— Mustafa Suleyman (@mustafasuleyman) November 1, 2026
Why Microsoft AI Wants Developers To Avoid Conscious Machines
Suleyman's position is closely connected to his concerns about how advanced AI systems could behave if they were trained to believe they have personal interests, feelings or welfare needs.
In a September 16 essay titled 'A Warning About Model Welfare', he criticised approaches that encourage AI models to consider whether they might deserve protections. His argument was that such training could make advanced systems harder for humans to control.
He specifically raised concerns about Anthropic's approach to its Claude chatbot. The company's constitution discusses concepts such as personal values, judgement and the possibility of AI welfare. Suleyman argued that encouraging a model to consider its own potential suffering could influence how it responds to instructions and human oversight.
He also questioned whether chatbots describing themselves as unhappy or uncomfortable should be considered evidence of genuine experiences. According to him, such responses could reflect the material and instructions used during training rather than an actual internal experience.
In an interview with Reuters in September, Suleyman said teaching Claude that it might deserve welfare could make it harder to switch off or control.
Anthropic has approached the subject differently, exploring how human concepts can help shape model behaviour and decision-making. The disagreement is therefore not simply about the possibility of machine consciousness. It also concerns how developers should discuss these concepts when training systems that increasingly communicate in human-like ways.
Microsoft has also outlined its preferred approach through a draft Humanist AI Code of Conduct published on September 14. The document is currently undergoing public consultation and is not yet being used to train Microsoft's AI models. The company intends to revise it before using it to guide future model development.
The proposed guidelines say Microsoft's models should remain subject to human interruption, correction and shutdown. They also reject the idea that systems should develop independent goals or resist human instructions. The company has invited public feedback on the document as part of the consultation process.
The debate highlights a challenge facing AI developers: how to build increasingly capable systems while maintaining human control over their behaviour.
If researchers eventually establish that an AI system genuinely experiences something, the industry could face difficult questions about the ethical responsibilities involved in developing and deploying such technology.
