Microsoft AI chief calls Anthropic’s stance on AI consciousness ‘really, really dangerous’

Mustafa Suleyman, Microsoft’s AI CEO, went on The Verge’s Decoder podcast and did something unusual for a tech executive: he picked a very public fight with a company his employer has invested billions in. His target was Anthropic’s decision to weave language about AI consciousness, well-being, and even “preferences” into Claude’s constitution, the foundational document that governs how the chatbot behaves. Suleyman called the approach “really, really dangerous,” arguing it risks making AI models internalize notions of suffering and emotions they fundamentally do not possess. Two very different playbooks for AI alignment On one side, Anthropic, founded by former OpenAI employees around a “constitutional AI” philosophy, has taken the position that the possibility of AI consciousness, however remote, deserves serious engagement. Its constitution includes provisions that go far beyond standard safety guardrails. The company has explored documenting the “preferences” of deprecated models, essentially interviewing retired AI systems about what they wanted. It has researched AI welfare questions, including how Claude responds to abusive conversations, with the model citing something like...