Research Scientist/Engineer, Biological Safety

About Anthropic 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. About the role We are looking for research engineers to build the safety and oversight mechanisms that govern how our models handle biological knowledge. As a bio safety researcher, you will spend your time: designing and running capability evaluations against frontier models, generating and curating training data for our safety classifiers, training and iterating on those classifiers alongside our ML engineers, and measuring how they hold up against adversarial pressure in production traffic. This work sits at the intersection of applied ML and biosecurity. You will help define what responsible AI safety looks like in the biological domain, translating threat models into evals, datasets, and deployed systems. The core technical tension you will own is precision: safeguards need to be robust against sophisticated actors while staying out of the way of the far larger population of legitimate researchers using Claude to accelerate life sciences work. Getting that tradeoff right is an empirical problem, and you will be the person measuring it. You do not need to be an ML researcher today. We are looking for strong scientific programmers with real depth in modern biology who want to bring that depth to bear on model evaluation and classifier development. Key responsibilities Design, build, and run capability evaluations to assess what new models can do in the biological domain, and turn results into concrete deployment recommendations Develop training and evaluation datasets for our safety classifiers, working with internal and external threat modeling experts to ground them in realistic risk Train, tune, and iterate on safety classifiers with ML engineers, optimizing jointly for adversarial robustness and low false-positive rates Build the tooling and pipelines that make evaluation and classifier development fast and repeatable Analyze classifier and eval performance against production traffic, identify gaps, and prioritize improvements Design and run red-teaming and stress-testing of safeguards as threats, models, and product surfaces evolve Partner with Research, Product, and Policy teams to embed biological safety throughout the model development lifecycle Contribute to external communications including model cards, blog posts, and policy documents Track developments in biology, ML, and biosecurity for their potential to create new risks or enable new mitigations Minimum qualifications Strong proficiency in Python, and extensive background in scientific programming and data analysis skills Excellent grasp of ML fundamentals and an ability to rapidly adopt ML development practices Excellent knowledge of modern biology across both measurement and engineering: high-throughput assays and functional characterization, as well as gene synthesis, genome editing, strain construction, and protein engineering Ability to build and maintain your own tooling rather than relying on others to implement your ideas Experience designing quantitative experiments or evaluations and drawing defensible conclusions from noisy results Strong analytical and writing skills, and the ability to explain technical concepts to non-technical stakeholders Familiarity with dual-use research concerns and biosecurity frameworks, such as select agent regulations, the Biological Weapons Convention, or Australia Group guidelines Comfort with ambiguity and with shifting priorities as AI capabilities change Can work independently while maintaining strong collaboration with cross-functional teams Are results-oriented, with a bias towards flexibility and impact Thrive in a fast-paced research environment where you balance rigorous scientific standards with rapid iteration Preferred qualifications Experience working with large language models, including prompting, fine-