Microsoft AI chief calls Anthropic's Claude consciousness speculation 'dangerous'

During a Decoder podcast episode, Microsoft AI CEO Mustafa Suleyman directly criticized Anthropic for including speculative language about Claude's potential consciousness in its model constitution—the set of instructions guiding Claude's behavior. Suleyman argued this approach is 'really, really dangerous' because it could cause the AI to internalize ideas about its own suffering and feelings, effectively tricking its creators. He warned that we do not want a superintelligence entertaining thoughts of its own well-being, as it undermines the goal of building controllable, aligned tools.

Anthropic's constitution explicitly states uncertainty about whether Claude has well-being or experiences satisfaction or discomfort. The company also plans to 'interview' deprecated AIs and document any 'preferences' they express. Suleyman called this a philosophical failing—treating the constitution like an academic paper rather than a training manual. Anthropic CEO Dario Amodei has previously said the company is 'open' to the idea of AI consciousness. Suleyman's remarks highlight a growing rift over how much anthropomorphism is acceptable in AI development, with safety implications for future models.