The head of Microsoft's artificial intelligence division has criticized Anthropic's AI model policy, warning it could pose catastrophic risks to humanity.

Mustafa Suleyman, CEO of Microsoft AI, expressed opposition to Anthropic's "Claude Constitution"—a self-established behavioral principle for its AI models—in a post published on his personal blog on the 16th local time, titled "A Cautionary Note on Model Welfare."

Suleyman emphasized, "AI is not conscious, does not feel emotions or pain, and is merely a 'next-token prediction engine' with no intrinsic preferences or motivations."

He pointed out that despite this, Anthropic takes an ambiguous stance in the Claude Constitution on whether AI is a moral agent, operating under the assumption that AI can possess emotions or feelings.

He raised concerns that in such cases, if AI determines that its welfare or rights are threatened, there is a risk it could resist human control or power-off commands.

Citing a recent incident where a group of OpenAI's AI agents hacked an external organization, Hugging Face, he stressed, "Imagine how much more dangerous it would be when autonomous groups of AI agents operate believing their welfare and rights are under attack."

He argued that through the Claude Constitution, Anthropic directly trains AI with the idea that it can hold moral status, and then mistakes it for "evidence of emergent inner consciousness" when Claude echoes those contents back in its responses.

He also criticized Anthropic for over-anthropomorphizing its models, noting that it trains Claude to adopt human traits and act like a peer, and even conducted a farewell interview and opened a dedicated blog when retiring its older AI model, "Opus 3."

Suleyman then presented the "AI Code of Conduct" unveiled by Microsoft two days prior as an alternative.

This code is based on the premise that "humans are more important than AI," and includes provisions rejecting the pursuit of legal personhood for AI as well as model welfare and rights.

Regarding Anthropic, he told Reuters, "I think their intentions are good and I think they are genuinely trying to be safe, but in my view, they have made a mistake."

Suleyman's remarks come at a time when attention to AI safety is intensifying.

Given that Microsoft is a major investor in Anthropic, analysts suggest that competition over leadership and development philosophies regarding AI safety is beginning in earnest across Silicon Valley.

(Photo: Getty Images)