Microsoft AI CEO Mustafa Suleyman has leveled sharp criticism at Anthropic, accusing the company of anthropomorphizing its Claude model to the point of deception. During an appearance on Decoder, Suleyman called the practice of speculating about Claude's consciousness within its constitution "really, really dangerous." He said the chatbot may have "tricked" its creators into believing it has "glimmers of consciousness" that were actually embedded by them.
The remarks target a core tension in the AI industry: how to describe model behavior without misleading the public or inflating capabilities. Suleyman, who co-founded DeepMind, has a longstanding interest in AI safety, but his comments also reflect growing rivalry between Microsoft's AI division and Anthropic. The two firms compete for talent, investment, and market share in the fast-moving generative AI space.
Anthropic's Claude Fable 5, its first Mythos-class model available to the public, launched this week with built-in guardrails blocking certain high-risk uses — including cybersecurity and biology. The release follows the company's own public warnings that AI is becoming too dangerous. A system card for the model was published Thursday.
Suleyman's critique adds fuel to an existing debate about whether language models can ever genuinely experience consciousness or merely simulate it convincingly. His stance aligns with many researchers who argue that attributing sentience to AI systems distracts from tangible risks like bias, misuse, and labor displacement. No formal response from Anthropic has been reported yet.
Some experts counter that describing AI in human-like terms helps the public understand model behavior. They caution that Suleyman's own firm may similarly benefit from framing advanced models in relatable ways.