What It’s Actually Like to Interview at Anthropic

The loop

1. Recruiter screen (30 min) — culture fit, motivation, salary expectations.
2. Technical phone screen (60 min) — ML fundamentals + one coding problem.
3. Take-home or live coding (~3 hours) — usually a small RL or RAG project.
4. On-site / virtual onsite (4-5 rounds):

• ML system design
• Coding (Python, no leetcode-style; more practical)
• Research discussion
• Behavioral / culture

What they actually evaluate

• Rigor: do you reason from first principles, or repeat blog-post talking points?
• Safety mindset: do you proactively think about misuse, failure modes, evaluation?
• Communication: can you explain technical decisions to a non-expert?

Tips that work

• Read the Anthropic blog before your interview. Their RSP, Constitutional AI, and interpretability papers come up indirectly.
• Have an opinion on alignment, but hold it loosely — they value updating on evidence.
• For system design: always discuss eval before scaling. "How would I know this is working?" is the right reflex.

Compensation (2026 estimates)

ML Engineer L4: $310K - $410K base + equity. SWE roles similar.

(Numbers from Levels.fyi composites; verify with your recruiter.)