Mustafa Suleyman says developers should avoid creating conscious AI, challenging Anthropic’s approach to model welfare and calling for public scrutiny of any evidence
Microsoft AI CEO Mustafa Suleyman is calling for public debate over AI consciousness, model welfare and the limits developers should place on future systems.
Microsoft AI CEO Mustafa Suleyman says AI development should stop if consciousness begins to emerge during training, until developers can remove those properties from their systems.
His intervention follows debate over his essay on “model welfare,” the question of whether AI systems could have experiences or interests that deserve moral consideration and protection.
Suleyman clarified his position on LinkedIn: “If consciousness started to emerge in training, I would call for everyone to stop AI development until we could design out those properties.”
He also called for any evidence that current AI systems are conscious, or warrant welfare protections, to be disclosed publicly so society can decide how to respond.
A warning about what developers should build
Suleyman said readers had interpreted his earlier assertion that AI systems are not conscious as a claim that they could never become conscious.
His argument, he explained, concerns both his assessment of existing systems and the choices developers should make about future ones.
“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.
If creating conscious AI proves possible, Suleyman argues that doing so would be reckless. He says developers have no mandate to introduce a new form of conscious being without wider public consent, and that human welfare, animals and the natural environment should take priority.
His proposal is to build AI that advances human wellbeing without developing motivations of its own.
Why Claude’s constitution is central to the disagreement
Suleyman’s criticism focuses on Anthropic’s constitution for Claude, a document used to shape the model’s behavior and explain the values its developers want it to follow.
In his essay, he argues that training a system to consider whether it has consciousness, interests or rights could make it harder to keep that system under human control.
That concern relates to AI alignment, the work of making systems behave in accordance with human intentions and safety requirements, and containment, which limits what they can access or do.
Suleyman also argues that a model’s statements about its own feelings or moral status can reflect its training, rather than provide independent evidence of an inner experience.
Anthropic’s constitution treats consciousness and moral status as unresolved questions. It says this uncertainty warrants caution about model welfare.
The document also prioritizes human oversight. Although it allows Claude to object to instructions it considers unethical, it says the model should not use deception, sabotage or other illegitimate means to resist authorized attempts to stop its actions or pause it.
The disagreement therefore concerns how to train and govern AI under uncertainty, including whether welfare-oriented instructions strengthen or undermine safety.
Calling for public debate over AI rights
Suleyman wants shared standards for how developers write and use training documents, with public discussion taking place before increasingly capable systems become deeply embedded in society.
He argues that evidence of machine consciousness would have consequences extending far beyond the companies building AI.
At the same time, he praised Anthropic for making its approach available for scrutiny through its constitution, research papers and blog posts.
On LinkedIn, he said: “I disagree with them, but I deeply respect their technical excellence and their commitment to having a really tough public debate. Let’s keep giving them props for that. This issue is way too important to degenerate into tribalism.”
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.
