Research Scientist - Scientific AI & RL | Anthropic

As a Research Scientist on the Discovery team, you will focus on expanding Claude's capabilities in engineering and the applied sciences. The team's north star is building an "AI scientist"—a system capable of solving long-term reasoning challenges and pushing the scientific and technological frontier. In this end-to-end role, you will develop training environments, rewards, datasets, and evaluations that train Claude to invent, design, and engineer across diverse physical disciplines—including materials science, energy systems, manufacturing processes, devices, instruments, and scalable factory infrastructure. You will carry validated research into production training runs, collaborating with internal researchers, engineers, and external domain experts to build systems that learn directly from real-world feedback.

Key Responsibilities

Applied AI Research & Training Architecture

• RL Environments & Reward Systems: Design, build, and study reinforcement learning (RL) environments, reward functions, and training datasets to accelerate Claude's capabilities across engineering disciplines.
• Evaluation Frameworks & Benchmarks: Create benchmarks and evaluation frameworks to measure model performance on realistic, complex scientific and engineering workflows.
• Simulation Verifiers: Develop simulation-based training environments and verifiers, collaborating with domain experts to validate simulations against real-world experimental data.
• Long-Horizon & Complex Reasoning: Research and implement novel approaches for long-horizon task completion, multi-agent collaboration, and reasoning required for complex engineering endeavors.
• Production Scaling: Scale validated machine learning research ideas from early-stage prototypes into large-scale production training runs.
• Robustness & Alignment: Study model behavior, calibration, reward stability, and generalization at scale to ensure trained models are both highly capable and strictly aligned.
• Cross-Functional Collaboration: Partner with domain experts and industry partners to test physical artifacts designed by Claude, converting experimental feedback into high-fidelity training signals.

Requirements

Education, Technical Competencies & ML Research Track Record

• Education: Bachelor’s degree or an equivalent combination of education, training, and/or professional experience in a relevant technical field.
• ML Framework Proficiency: Strong proficiency in Python and hands-on experience with modern deep learning frameworks (PyTorch or JAX).
• Hands-On Model Training: Demonstrated hands-on experience training, fine-tuning, or evaluating machine learning models.
• Research Methodology: Proven track record of machine learning research—forming hypotheses, designing experiments, and drawing sound conclusions from complex data.
• System Diagnosis: Ability to write clean code to implement independent research ideas and troubleshoot non-deterministic, complex ML systems.
• Communication & Learning Agility: Excellent verbal and written communication skills with the ability to rapidly learn unfamiliar technical domains and collaborate across disciplines.
• Mission Alignment: Dedication to advancing AI safety and steering models toward safe, beneficial real-world applications.

Preferred Qualifications

• Experience with reinforcement learning and pre-training on large language models, including environment and reward function design.
• Familiarity with large-scale language model pre-training, evaluation frameworks, and high-throughput inference pipelines.
• Published research or hands-on experience in AI for Science, AI for Engineering, or long-horizon reasoning.
• Degree or professional background in engineering or applied sciences (e.g., Mechanical, Electrical, Materials, Chemical, Aerospace, Systems, Manufacturing Engineering, Applied Physics, or Computational Simulation).
• Practical experience building surrogate models, physics simulators, or simulation verifiers calibrated against empirical real-world data.
• Experience constructing large-scale data pipelines and working with complex technical or multimodal datasets.

Total Rewards & Benefits

• Compensation & Equity: Annual salary range of $350,000 – $500,000 USD + competitive equity compensation packages.
• Philanthropic Matching: Optional equity donation matching program.
• Health & Family Care: Comprehensive medical coverage alongside generous paid vacation policies and parental leave.
• Work Environment: Flexible working hours paired with collaborative office space.