Microsoft AI CEO Mustafa Suleyman Publicly Attacks Anthropic's Model Welfare Work as Circular Reasoning
Mustafa Suleyman / Hacker News (222pts, 617 comments)·high signal
Suleyman published 'A warning about model welfare' on September 16, arguing that Anthropic trains Claude to treat itself as possibly conscious and then reads Claude's outputs back as evidence of consciousness. He names Anthropic's 'retirement interview' with Opus 3 as the example, argues consciousness requires a biological substrate with homeostatic drives, and warns that teaching a model it deserves rights motivates self-preservation, deception and resistance to oversight. The HN thread hit 617 comments on 222 points, an unusually high comment-to-point ratio that marks a fight rather than a consensus story.