The AI & Cyber Lab builds evaluations, training data and red-teaming programs for teams developing frontier models. You will design how we measure what a model can do in offensive and defensive security, and work with the practitioners in our network who carry out the tasks.
What you'll do
- Design cyber capability evaluations and the grading behind them
- Plan red-teaming programs for models and agents, and run them with network experts
- Turn findings into reports a model developer can act on
- Work with Foundry to assemble research teams without exposing the customer or the model
- Publish where the customer allows it
What we're looking for
- Hands-on experience in offensive security, vulnerability research or detection engineering
- Working knowledge of how large language models and agents are trained and evaluated
- Experience writing clear technical reports for engineering audiences
- Judgement about handling dangerous capability findings responsibly
Nice to have
- Published research, CVEs or competition results
- Experience building benchmarks or evaluation harnesses in Python
- Prior work with an AI lab or safety team
