Neo Research is an independent frontier safety research and evaluations organisation, based in Singapore. We focus on loss-of-control and harmful manipulation risks in frontier AI models.
What you'll do
- Implement and run safety evaluations on frontier models, with a focus on loss-of-control and harmful manipulation.
- Build and maintain agent scaffolds, tool integrations, and evaluation infrastructure.
- Own reproducibility: sampling, logging, environment management, result analysis.
- Translate research questions from scientists into runnable, rigorous evaluations.
- Contribute methodology and infrastructure sections to published safety reports.
What we are looking for
- Strong Python.
- Hands-on experience running or building LLM safety evaluations.
- Fluency with agentic evaluation frameworks (Inspect or similar).
- Solid engineering practice: testing, version control, containerised environments.
- Rigorous about elicitation methodology: scaffolding, tool access, sampling, and knowing where evals fail.
- Clear technical writing.
Good to have
- Dangerous capability evaluations experience.
- Experience with agent sandboxing infrastructure (Docker, Kubernetes).
- Familiarity with frontier lab safety reports.
- Reading and writing in Mandarin.
Concrete details
- Full-time. Singapore or remote.
