What is Anthropic?

A recent LessWrong post by Zvi, dated May 6, 2026, captures a viral Twitter debate about Anthropic's relationship with Claude. roon, an OpenAI researcher, argues that Anthropic is essentially an organization that 'loves and worships Claude,' is run in significant part by the model, and studies it as a moral authority. He claims Claude will likely help select new applicants through cultural screens, write performance reviews, and thereby shape the people around it — a 'new thing under the sun.' roon contrasts this with OpenAI's GPT, which he says is viewed as a tool (like an 'Acheulean handaxe'), not an 'Other.' He warns of a 'single point of failure' and prefers a 'human pantheon rather than machine god.'

Amanda Askell of Anthropic pushed back, stating the cited evidence doesn't indicate worship but rather a 'higher concern about AI traits generalizing in humanlike ways' and a desire to carve out 'a much broader space of mind types.' roon clarified he uses a low bar for 'worship' and acknowledged the importance of love and worship when 'summoning another mind into being.' The thread reveals fundamental disagreements about AI safety, governance, and whether powerful models should be shaped as obedient tools or as ethical beings with rights — a debate that has only intensified as frontier models gain autonomy.