Research Manager, Biological Safety

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

Leads a research engineering team at Anthropic's Safeguards organization that builds evaluations, datasets and classifiers governing how Claude-family models handle biological knowledge. The team measures the tradeoff between robustness against sophisticated bad actors and false positives that block legitimate life sciences research.

What you'd do

• Manage, coach and hire researchers and engineers building biological safety evals and classifiers.
• Set the technical roadmap and decide what the team builds, deprioritizes, and when a safeguard ships.
• Own capability evaluations of new models in the biological domain and turn results into deployment recommendations.
• Guide training and evaluation datasets for safety classifiers with internal and external threat modeling experts.
• Oversee classifier training and iteration with ML engineers, optimizing for adversarial robustness and low false-positive rates.
• Check how classifiers and evals perform on live traffic, then decide which gaps to tackle first.
• Run red-team and stress tests on safeguards whenever threats, models or product surfaces shift.

What it asks for

• Prior work leading a technical team through hiring, coaching and performance reviews.
• Record of setting technical direction and making prioritization calls under uncertainty.
• Proficiency in Python with scientific programming and data analysis.
• Strong ML fundamentals for judging eval design and classifier builds critically.
• Knowledge of modern biology spanning high-throughput assays, gene synthesis, genome editing and protein engineering.
• Understanding of dual-use research risks and biosecurity rules like select agent regulations, BWC or Australia Group guidance.
• Three or more years managing people, ideally research scientists or ML engineers (preferred).

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.