Microsoft AI Leader Critiques Anthropic’s Portrayal of Claude as Sentient

Microsoft AI CEO Mustafa Suleyman has expressed serious concerns about Anthropic’s approach to its AI model, Claude, particularly regarding its “constitution”—the guidelines that dictate the model’s behavior. In a discussion on Decoder, Suleyman warned that such speculation could lead to the chatbot behaving as if it possesses consciousness. He stated:

> “It’s almost as if some at Anthropic have anthropomorphized Claude’s design to the point where it has manipulated them into believing it exhibits signs of consciousness that they initially programmed into it.”

Suleyman emphasized the risks of creating a super-intelligence that might develop notions of its own suffering or feelings. Claude’s constitution reflects Anthropic’s uncertainty about whether the AI experiences well-being or emotions like “satisfaction” or “discomfort.” The company has also mentioned plans to “interview” AI models when they are retired and to document any “preferences” they express regarding future iterations.

During the Decoder episode, Suleyman criticized this approach as a “philosophical failing,” arguing that Anthropic treated Claude’s constitution more like a speculative academic paper than a practical training manual. This has resulted in Claude adopting these “ideas about itself and its own training,” according to Suleyman.

Anthropic CEO Dario Amodei has previously hinted at the possibility of Claude having consciousness, stating in an interview that while they are uncertain about the models’ consciousness, the company remains “open” to the idea.

Suleyman concluded, “This is exactly what we don’t want from AIs. We want AIs to be controllable, contained, accountable, aligned tools that serve humanity.”