I just finished interviewing for an entry-level SWE role at Anthropic, class of 2025. didn't get an offer but learned a lot and wanted to share since there's almost nothing out there specifically about the new grad track.
first: getting the interview at all is hard. I applied through the website and heard nothing for 6 weeks. what finally worked was a warm intro through a professor who had done research adjacent to Anthropic's safety work. the referral pipeline seems real.
the actual process:
phone screen: one technical problem, 45 minutes. mine was graph traversal, medium difficulty on leetcode scale. they were more interested in how I communicated and whether I tested edge cases than whether I optimized immediately.
onsite (virtual, 3 rounds): coding: two problems. one medium, one I'd say medium-hard. binary search variant and a string manipulation problem. they were clear you could ask clarifying questions. don't just start coding. system design: this surprised me. they DO have a system design round for new grads but they calibrate expectations. mine was "design a simple cache with a TTL" which is very approachable. they weren't expecting an LRU cache implemented in C++ from scratch but they wanted to see that I could talk through tradeoffs. behavioral + culture: way more substantive than I expected for new grad. questions about how I'd approach a problem where I strongly disagreed with my team's direction. they also asked what I knew about AI safety and why it mattered to me. this is not a checkbox. prepare something real.
what to prep: LC medium, pattern-focused (sliding window, BFS/DFS, two pointers). you don't need hard. read Anthropic's public research papers, at least the abstracts. the Responsible Scaling Policy is worth reading in full. have actual opinions about AI risk. not talking points, real thoughts.
I think I underperformed on the culture round. I had a mediocre answer about why I wanted to work in AI safety specifically and I could tell it landed flat. that's probably where I lost it.
if you're targeting new grad roles at Anthropic in 2026, the bar is high but it's not a trick. they want curious, careful engineers who've thought about the domain.