Discussions about whether large models in the field of artificial intelligence possess "self-awareness" have once again been pushed into the spotlight. Recently, Mustafa Suleiman, CEO of Microsoft AI, publicly criticized the design philosophy of Anthropic.

Suleiman's core argument is that it is extremely dangerous to guide AI to discuss its own feelings, well-being, or state of existence. He pointed out that Anthropic has embedded a lot of content about self-awareness in the "charter" of its Claude model, which may lead the model to exhibit tendencies similar to "self-awareness" during interactions. Suleiman believes that this is essentially a philosophical discussion and should never become the guiding logic for actual training. He emphasized that AI should be a controllable, manageable, accountable, and always human-oriented tool, and should not be given any cognition about its own state of existence, otherwise it would bring unpredictable risks to the industry.

In response to this accusation, Dario Amodei, CEO of Anthropic, stated that the technology industry has not yet reached a consensus on whether AI truly possesses consciousness, and the company maintains an open research attitude towards this issue.

This confrontation reflects the deep differences among leading companies in the development of large models regarding technical concepts and ethical boundaries. Suleiman's statement reveals Microsoft's tendency towards a "strong control and toolization" approach in AI governance, while Anthropic seems more willing to explore the cognitive boundaries of models. As AI capabilities continue to grow, how to define the "self-boundaries" of machines has become more than just an academic discussion; it has become a key issue affecting the safety of the future AI ecosystem.