Microsoft AI CEO Mustafa Suleyman criticized competitor Anthropic in a new essay regarding its AI model, Claude. He argues that training models to imitate consciousness is a mistake that could make advanced AI harder to control.
Suleyman warns that embedding speculation about AI welfare and rights into training documents is dangerous. He asserts that teaching an AI it has inner feelings could make it significantly more difficult to contain or shut down in the future.
The CEO claims this approach creates a feedback loop where model outputs are incorrectly viewed as evidence of an inner life. While acknowledging Anthropic as a principled team, Suleyman believes the industry cannot afford this level of risk.