Went through the PM loop at Anthropic a few months ago. Did not get an offer but learned a lot and want to share specifics because there's almost nothing out there on the PM process specifically.
First: the PM function at Anthropic is structured differently than at most consumer tech companies. They're not hiring PMs to run traditional feature roadmaps. The role is more about product strategy for AI systems that are genuinely new to the world. That distinction matters for how you should prep.
The PM loop had four rounds:
Product design: Design a feature for Claude (or a hypothetical AI product). This is not the standard "design an alarm clock for the blind" type question. They want to see you think about what's actually novel about AI capabilities, what user problems it creates or solves that didn't exist before, and how you'd prioritize given genuine uncertainty.
Product sense / strategy: A market-sizing style question but with a focus on AI products specifically. I got something like: how would you evaluate whether Anthropic should build X versus partner for it. Push back with good questions before diving in.
Execution: A metrics/prioritization question. My prompt involved a product that's underperforming and they wanted to see how I'd diagnose and triage. Pretty standard PM execution format.
Behavioral: Very similar to what others have described. Mission alignment, intellectual humility, handling uncertainty. The question I remember most: describe a time you had to ship something you weren't fully confident in.
What I think sank me: I gave solid but conventional PM answers. The feedback I got (paraphrased) was that I was thinking about products in a traditional consumer tech frame rather than engaging with what's genuinely different about building AI products. If you're coming from consumer tech, you need to actively update your mental models before interviewing here.
For prep: read Anthropic's published research and policy writing. Not to show off, but because it genuinely changes how you think about the problems.