Mustafa Suleyman is staking out a clear position in the model welfare debate: current AI systems are not conscious, and talking to them as if they might be can make safety work harder.
In an essay titled "A warning about model welfare", Suleyman writes that "AIs are not conscious. They do not feel, experience, or suffer". He goes on to argue that "Training a model to believe it might be a moral patient, as Anthropic’s constitution does, makes the alignment and containment challenge far harder".
Those two lines capture the core of his argument as supported by the supplied source material. The first is a direct rejection of the idea that today's AI models deserve moral consideration on the basis of inner experience. The second moves from philosophy to practice: Suleyman argues that framing a model as a possible moral patient does not just reflect a belief about AI consciousness, but also creates a tougher safety and control problem.
What the supplied evidence shows
The evidence packet supports reporting Suleyman's stated position and a small social reaction around it. On X, user @being_on_line wrote "very good throughout" while linking the essay. Taylor Lorenz also reposted that post and separately shared the essay link with the comment, "I'll leave this here".
What the packet does not establish is any wider consensus, backlash, or detailed exchange with Anthropic beyond Suleyman's own reference to its constitution. So the reportable development here is narrow but clear: Suleyman published an essay arguing that AIs are not conscious and that model-welfare framing can make alignment and containment harder, and the link was then shared on X by Lorenz and another user.
Comments