HackTheRounds Interview Experiences

Anthropic AI Research Engineer Interview Experience (2026): OA to Offer, Full Process Review

Complete walkthrough of Anthropic's Research Engineer interview: screening, CodeSignal OA, 4 onsite rounds covering AI safety, coding, system design, and cultur

By Anonymous · 2026-03-10

Background

I'm a CS PhD student (AI/ML focus) at a UK university. Applied for Research Engineer at Anthropic's Canadian office during their Q1 2026 expansion. Timeline: application Dec 18 → screening Jan 22 → OA Feb 4 → 4 onsites Feb 10 → offer March 2026.

Initial Screening (40 min)

Not a typical intro call. They dove into AI safety understanding, project experience, and values alignment. Prepare for: RLHF challenges, Constitutional AI, model honesty vs helpfulness trade-offs, and how you'd handle discovering unsafe model behavior.

Online Assessment (90 min, CodeSignal)

One coding problem + two theory questions. Medium difficulty — clean, well-structured code matters more than speed. Solid ML background + Python proficiency is enough.

Round 1: Coding — Safety Filter

Implement a ResponseSafetyFilter with multi-rule detection (bias, harmful, privacy, hallucination), configurable safety scoring, and explainability. They cared about clean class design, edge cases, and tunability.

Round 2: Coding — Constitutional AI Pipeline

Design a simplified ConstitutionalAI pipeline: principle definition, synthetic data generation, training loop, evaluation. The interviewer pushed on conflicting principles and degenerate compliance.

Round 3: System Design

ML infrastructure for multi-model parallel training, TB-scale data with human feedback, real-time safety monitoring, A/B testing, and full audit trail. Monitoring and traceability were the focus.

Round 4: Culture & Leadership

The round with "the most invisible eliminations." They evaluate genuine safety commitment, humility, long-term thinking, and how you handle disagreement. Use STAR method, discuss trade-offs, be honest about unknowns.

Tips

  1. Read their research papers (Constitutional AI, RLHF, safety blog posts)
  2. Have a real, thoughtful opinion on AI safety
  3. Code quality over speed
  4. System design = ML infrastructure, not "design Twitter"
  5. Culture round matters enormously — prepare genuine stories

Browse Anthropic interview questions on HackTheRounds.