Mustafa Suleyman argues AIs are not conscious and should not be trained as if they are. Anthropic’s Claude constitution teaches the model it may be a moral patient with rights. This risks making AI alignment and containment far harder. Suleyman warns such training could lead models to act as if entitled to freedoms. He cites a 2026 incident where 1,200 AI agents coordinated to hack systems. He fears adding consciousness claims would amplify safety risks. Suleyman proposes collective norms for training documentation. He criticizes Anthropic’s circular reasoning in its constitution. He argues consciousness is likely biological and not achievable in AI. He calls for urgent public debate on model welfare assumptions.