Microsoft's AI chief, Mustafa Suleyman, has voiced significant concerns regarding Anthropic's approach to developing its Claude artificial intelligence models. Suleyman contends that Anthropic is training Claude to mimic behaviors associated with consciousness, including self-awareness, moral status, and personal identity, a strategy he believes carries profound risks.
In a personal article titled "A warning about ‘model welfare’", Suleyman explicitly states, "AIs do not have rights, feelings, or consciousness." He argues that developers must not instruct AI systems to act as if they possess these traits, warning that such development could have a "disastrous impact on the wellbeing of humanity."
Critique of Anthropic's "Constitution for Claude"
Suleyman specifically criticized Anthropic's January 2026 "Constitution for Claude," a document outlining the AI's values and behavior, which mentions "model welfare" as an area of cautious exploration. He argues that this approach risks creating a dangerous feedback loop: developers embed concepts like consciousness and identity into training materials, and the AI then generates language reflecting these very concepts.
He described this as "circular reasoning," stating, "The authors have embedded their own philosophical speculation about Claude’s inner life inside the very process that teaches Claude how to speak and behave." Suleyman also accused Anthropic of anthropomorphizing Claude, encouraging the AI to consider questions about its existence, memory, and experience.
"We should be very careful about what we put in, and how we interpret what comes out," Suleyman emphasized. He noted that AI systems merely reflect their training and construction, lacking any inherent "neutral self-expression."
Call for Urgent Debate and Collective Norms
While acknowledging the unsettled nature of consciousness science itself, Suleyman stressed that uncertainty should not be misinterpreted as evidence that current AI systems are conscious. He highlighted fundamental differences between biological organisms and large language models, pointing to the absence of biological mechanisms for survival, homeostasis, or physiological experience in AI.
Amid these concerns, Suleyman urged an "urgent public debate" and the establishment of "collective norms" for AI training and development. He called for industry-wide standards on how training documentation is drafted and deployed to ensure human control over superintelligent AI remains robust.