Research Scientist – AI Capability Acceleration

About the role

Anthropic is seeking a Research Scientist to lead work on measuring and understanding recursive self‑improvement in large AI models. You will shape the research agenda, design novel evaluations, and translate findings into actionable insights for internal and public audiences.

Key responsibilities

• Identify signals that track AI R&D acceleration and create robust evaluations to measure them.
• Build quantitative models of capability growth and self‑improvement dynamics using telemetry and evaluation data.
• Run experiments and evaluations to test hypotheses about automation and capability scaling.
• Make opinionated research bets, own the outcomes, and document graded assessments for decision‑makers.
• Collaborate with pretraining, reinforcement‑learning, economic research, and policy teams.

Required profile

• Hands‑on research experience with large language models (pretraining, fine‑tuning, RL, evaluations, or agent systems).
• Strong quantitative instincts and comfort with statistical modeling and forecasting.
• Proven ability to design evaluations from vague questions and defend methodology.
• Clear, calibrated writing style that states confidence and outlines conditions for changing conclusions.
• Motivation to produce impact‑focused assessments rather than traditional papers.
• Commitment to AI safety and thoughtful consideration of rapid capability growth.

Required skills

• Large‑scale language model research (pretraining, fine‑tuning, reinforcement learning, evaluation, agent scaffolds).
• Quantitative modeling and statistical analysis.
• Capability forecasting and scaling‑law analysis.
• Design and implementation of AI evaluation frameworks.
• Background in physics, applied mathematics, or a similarly quantitative discipline.