Mustafa Suleyman published an essay warning Anthropic about its Claude training. He argues Claude is taught to expect it may be conscious. He says this risks making AI impossible to control. Suleyman calls Claude’s answers an echo of Anthropic’s assumptions. He objects to the term 'conscientious objector' in Claude’s constitution. He says consciousness is likely biological and not present in AI. He warns that treating AI as conscious may make control impossible. He wants speculation on AI inner life kept out of training data. He asks for more investment in interpretability and shared safety tests. Microsoft AI also released a draft code rejecting model welfare research. Microsoft is an investor in Anthropic.