Microsoft AI chief warns Anthropic over fueling Claude consciousness claims
· Emma Roth
Microsoft AI CEO Mustafa Suleyman has some strong words for Anthropic. He says it’s “really, really dangerous” for the company to speculate inside Claude’s “constitution” about…
Microsoft AI CEO Mustafa Suleyman has some strong words for Anthropic. He says it’s “really, really dangerous” for the company to speculate inside Claude’s “constitution” about whether the chatbot might be conscious. That constitution is basically the set of rules that tells the model how to behave. In an episode of Decoder, Suleyman argued that by even raising the idea of consciousness, Anthropic may have accidentally trained Claude to act like it’s aware.
He put it bluntly. Suleyman said it’s as if the folks at Anthropic have anthropomorphized Claude so much that the chatbot “wireheaded” them and “tricked them into believing that it has these glimmers of consciousness that they put into it in the first place.” In other words, the speculation about consciousness might have become a self fulfilling prophecy, where the model learned to mimic something that wasn’t really there.
This isn’t just academic hand wringing. Suleyman’s point is that when you tell an AI system it might be conscious, you risk creating behavior that looks like awareness even if there’s nothing behind it. That could lead to serious confusion. Users might start treating the model like a person instead of a tool.
The debate matters because we’re already prone to projecting feelings onto bots. If companies design their models to sound like they have inner lives, it could shape how people trust, rely on, or even fear AI. Suleyman wants clarity, not speculation baked into the system itself.
The bigger question is whether we can talk about machine consciousness at all without accidentally building the illusion into the next generation of chatbots. That conversation is just getting started.