Research Scientist/Engineer, Biological Safety

Written by the T&S Network from Anthropic's posting.

Builds the safety and oversight mechanisms that govern how a frontier AI model handles biological knowledge. The work sits at the intersection of applied ML and biosecurity, covering capability evals, safety classifiers, and red-teaming of safeguards. The core problem is balancing robust protection against sophisticated bad actors with not blocking the much larger population of legitimate life sciences researchers.

What you'd do

• Design and run capability evaluations of new models in the biological domain and turn results into deployment recommendations
• Build training and evaluation datasets for safety classifiers with internal and external threat modeling experts
• Iterate on safety classifiers with ML engineers, balancing resistance to attacks against wrongly flagging benign use
• Measure classifier and eval results on live traffic, find weak spots, and rank fixes
• Probe safeguards with adversarial testing as threats and product surfaces shift
• Work with Research, Product, and Policy to build biological safety into model development
• Contribute to model cards, blog posts, and policy documents

What it asks for

• Strong Python and scientific programming with data analysis skills
• Excellent grasp of ML fundamentals and ability to adopt ML practices quickly
• Excellent knowledge of modern biology, including high-throughput assays, gene synthesis, genome editing, and protein engineering
• Can build and upkeep your own tooling without waiting on others
• Experience designing quantitative experiments and drawing conclusions from noisy results
• Working knowledge of dual-use and biosecurity rules, like select agent regulations, the BWC, or Australia Group lists
• Bachelor's degree or equivalent experience in a relevant field

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Anthropicnorth_east before you apply. If we got something wrong, report this listing.