Research Scientist/Engineer (Science of Scheming)
Apollo Research · London & San Francisco
CHECKED 8H AGO
Research instituteonsite
At a glance
Highlights
- Build a new empirical science of scheming to predict risks as models scale.
- Study dynamics behind reward-seeking and evaluation awareness in frontier models.
- Develop novel evaluations built to scale to highly evaluation-aware AI systems.
Ideal candidate
Fast-moving empirical researcher fascinated by AI cognition and scheming risks, driven to turn vague ideas into concrete experiments and relentlessly push progress forward.
Requirements
- Able to design and execute fast-paced empirical experiments
- Deep conceptual understanding of AI scheming and relevant literature
- Strong software engineering skills with Python stack
- Intense interest in AI progress and AI cognition
- Hands-on experience training LLMs via reinforcement learning
- Strong analytical skills with quantitative modeling and noisy data
- hiring process
- Conducting interviews actively to fill role as soon as suitable candidate found
Lab profile
About Apollo Research
AI safety organisation detecting and mitigating scheming in frontier AI systems and coding agents.
What they build
- Watcher – coding agent security monitoring product
Known for
- Pre-deployment testing with frontier AI labs
- Scheming detection evaluations
- Science of scheming and mitigations research
- Production monitoring of billions of agent tokens monthly