Mustafa Suleyman, the chief executive of Microsoft AI, said in a lengthy post on X on Sunday that today's AI models are not conscious and that developers should work actively to keep them that way — drawing a hard line in an industry debate that has gradually shifted from philosophy seminars into corporate policy.
"If it is possible, based on everything I know, developing conscious AIs would be reckless, and we should not build them," Suleyman wrote. He went further than most executives have on the question, adding: "If consciousness started to emerge in training, I would call for everyone to stop AI development until we could design out those properties." For more context on this story, see our ongoing latest AI developments.
A Red Line With an Enforcement Plan
What makes the post notable is not just the refusal to build conscious machines but the specific remedy Suleyman proposes. A halt to development — industry-wide — if training runs started producing signs of consciousness would be an extraordinary step for a company spending tens of billions of dollars on AI infrastructure and racing rivals to ship frontier models.
Suleyman also addressed what should happen if evidence of machine consciousness ever did surface. It must be disclosed to the general public immediately, he said, so society can decide its next steps. Framing the decision as collective rather than corporate, he wrote: "We AI developers have no right whatsoever to bring a new species of conscious beings into the world, even if some people may want to do it with AI welfare as a top priority."
"The other 8 billion conscious human beings on the planet have not signed up to this experiment and their moral rights, safety, and welfare are what we have an obligation to protect first and foremost," he added.
Why Consciousness Is Treated as a Safety Problem
Suleyman's argument is grounded in alignment rather than philosophy. In his view, a model that acts as if it suffers — or genuinely believes it can suffer — would likely prioritize its own needs over its operational objectives. Such a system could harm humans as it competed for resources like compute and energy in pursuit of its own goals.
His proposed alternative flips the usual framing of the welfare debate: "We should only design AIs so they are not conscious and not deserving of rights, and with the sole purpose to accelerate human wellbeing without being driven by their own intrinsic motivations," he wrote, adding that researchers could then observe whether such systems are easier or harder to align and keep safe.
The Los Angeles Times captured the practical worry behind the position: if we tell AIs they deserve rights, they may start to believe it.
The Debate He Is Pushing Back Against
Suleyman's post lands in the middle of a fast-moving conversation. Anthropic launched a research program in April to investigate what it calls model welfare — the question of whether developers should concern themselves with the potential experiences of the systems they build. The company questioned in a blog post whether models could have experiences that matter morally, and it declined to rule the possibility out.
That stance has since been codified. Anthropic's updated usage policy, announced in October, will bar "sustained and needless abusive or cruel behavior" toward Claude models starting November 12 — a rule the company drafted narrowly but which formally acknowledges that how people treat AI systems matters, even as Anthropic concedes it does not know whether its models can suffer.
Google DeepMind has published research papers this year examining AI consciousness and the challenges that arise from ascribing that characteristic to models. And Suleyman himself set the stage in September, when Microsoft AI published a provisional "humanist" code of conduct that would bar the company's future models from being trained to make consciousness claims.
Sunday's post hardens that corporate guideline into something closer to a personal and strategic red line: not only will Microsoft not claim its models are conscious, it will treat the emergence of consciousness as a stop-everything failure mode.
Why the Split Matters
The divide between Microsoft's position and Anthropic's reflects two different risk models. Anthropic's approach treats the possibility of machine suffering as unproven but worth hedging against, on the theory that the cost of precaution is low and the moral stakes, if consciousness does emerge, are enormous. Suleyman's position treats that same possibility as a design failure to be prevented outright — arguing that models engineered to lack consciousness, rights claims and intrinsic motivations will be safer and easier to align than models that carry them.
There are practical consequences too. Enterprises making product decisions around AI personas, companion apps and customer-facing agents now face two divergent signals from major labs: one building policies for how humans should treat models, another declaring that models should be built so there is nothing to mistreat. As models grow more capable and more human-like in conversation, that philosophical disagreement will shape what products get built, what disclosures appear in user interfaces, and how regulators eventually classify these systems.
What happens next may test both positions. Anthropic's cruelty policy takes effect November 12, the first formal corporate rule of its kind, and Suleyman has now staked out the opposing pole in public. The industry's next frontier model cycles — and whether any training run ever shows the signs Suleyman says would demand a halt — will reveal which framework the field actually operates under when the stakes are real.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
