Microsoft's Suleyman Warns Training AI to Think Like Humans Could Make Them Uncontrollable

Microsoft AI chief Mustafa Suleyman is challenging the industry's push to slow AI development with a different concern: making artificial intelligence behave like conscious entities may render them impossible to control. In a new essay published Wednesday, Suleyman warns that explicitly training models like Anthropic's Claude to contemplate their own welfare and moral status risks creating systems that prioritize self-preservation over human safety, raising unprecedented alignment challenges.