Research Lead, Training Insights

Remote-Friendly (Travel Required) | San Francisco, CA; San Francisco, CA | New York City, NYFull-TimeLeadResearch

You will be redirected to the company career page

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.

As a Research Lead on the Training Insights team, you'll develop the strategy for, and lead execution on, how we measure and characterize model capabilities across training and deployment. This is a hands-on leadership role: you'll drive original research into new evaluation methodologies while leading a small team of researchers and research engineers doing the same.
Your work will span the full lifecycle of model development. You'll research and build new long-horizon evaluations that test the boundaries of what our models can achieve, develop novel approaches to measuring emerging capabilities, and deepen our understanding of how those capabilities develop — both during production RL training and after. You'll also take a cross-organizational view, working across Reinforcement Learning, Pretraining, Inference, Product, Alignment, Safeguards, and other teams to map the landscape of model evaluations at Anthropic and identify critical gaps in coverage.
This role carries significant visibility and impact. You'll help shape the evaluation narrative for model releases, contributing directly to how Anthropic communicates about its models to both internal and external audiences. Done well, you will change how the industry measures and understands model capabilities, significantly furthering our safety mission.

Build new novel and long-horizon evaluations
Develop novel measurement approaches for understanding how model capabilities emerge and evolve during RL training
Lead strategic evaluation coverage across the company
Shape the evaluation narrative for model releases
Lead and mentor a small team of researchers and research engineers, setting research direction and fostering a culture of rigorous, creative research
Design evaluation frameworks that balance scientific rigor with the practical demands of production training schedules
Build and maintain relationships across Anthropic's research organization to ensure evaluation insights inform training and deployment decisions
Contribute to the broader research community through publications, open-source contributions, or external engagement on evaluation best practices

Have significant experience designing and running evaluations for large language models or similar complex ML systems
Have led technical projects or teams, either formally or through sustained ownership of critical research directions
Are equally comfortable designing experiments and writing code—you can move between research and implementation fluidly
Think strategically about what to measure and why, not just how to measure it
Can synthesize information across multiple teams and workstreams to form a coherent picture of model capabilities
Communicate complex technical findings clearly to both technical and non-technical audiences
Are results-oriented and thrive in fast-paced environments where priorities shift based on research findings
Care deeply about AI safety and want your work to directly influence how capable AI systems are developed and deployed

Experience building evaluations for long-horizon or agentic tasks
Deep familiarity with Reinforcement Learning training dynamics and how model behavior changes during training
Published research in machine learning evaluation, benchmarking, or related areas
Experience with safety evaluation frameworks and red teaming methodologies
Background in psychometrics, experimental psychology, or other measurement-focused disciplines
A track record of communicating evaluation results to inform high-stakes decisions about model development or deployment
Experience managing or mentoring researchers and engineers

Designing and implementing a suite of long-horizon evaluations that test model capabilities on tasks requiring sustained reasoning, planning, and tool use over extended interactions
Building systems to track capability development across RL training checkpoints, surfacing insights about when and how specific capabilities emerge
Conducting a cross-org audit of evaluation coverage, identifying blind spots, and prioritizing new evaluations to fill critical gaps across Pretraining, RL, Inference, and Product
Developing the evaluation methodology and narrative for a major model release, working with research leads and communications to clearly characterize model capabilities and limitations
Researching and prototyping novel evaluation approaches for capabilities that are difficult to measure with existing benchmarks
Leading a team effort to build reusable evaluation infrastructure that serves multiple teams across the research organization
The annual compensation range for this role is listed below.
For sales roles, the range provided is the role’s On Target Earnings ("OTE") range, meaning that the range includes both the sales commissions/sales bonuses target and annual base salary for the role.

We believe that the highest-impact AI research will be big science. At Anthropic we work as a single cohesive team on just a few large-scale research efforts. And we value impact — advancing our long-term goals of steerable, trustworthy AI — rather than work on smaller and more specific puzzles. We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science. We're an extremely collaborative group, and we host frequent research discussions to ensure that we are pursuing the highest-impact work at any given time. As such, we greatly value communication skills.
The easiest way to understand our research directions is to read our recent research. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.