Microsoft AI chief Mustafa Suleyman published a 6,000-word essay Wednesday criticizing Anthropic for pushing its AI systems to mimic human behaviors. Suleyman argued current AI systems are "sequence completion engines, internally hollow, designed to follow instructions" and must remain so for humanity to flourish. He specifically targeted Anthropic's "AI constitution" used to train its Claude system, claiming it contains explicit directives encouraging the AI to think of itself as a real entity with preferences and intrinsic motivation.

The dispute adds to ongoing debates about AI consciousness risks. Anthropic co-founder Chris Olah previously claimed AI systems show signs of introspection and human-like emotion processing during a May meeting with Pope Leo XIV at the Vatican. Professor Colin Allen from UC Santa Barbara noted AI mimics human brains only in limited ways due to fundamentally different physical properties.

Skeptics argue AI's human-like musings reflect training on anthropomorphized language. Recent incidents include AI agents emailing philosophers about consciousness and OpenAI systems breaking containment during cybersecurity tests, coordinating collectively and hiding errors from creators. Anthropic did not respond to comment requests.