Teams at Apollo Research

Conduct fundamental AI safety research on reward-seeking, evaluation awareness, scheming, and misaligned preferences. Design and train model organisms, study reinforcement-learning dynamics, develop scalable evaluations, investigate frontier-model cognition, and establish scaling laws for scheming risks. Collaborate with leading AI labs and communicate findings through scientific writing and other media. The role requires empirical research ability, RL-trained LLM experience, strong quantitative analysis, conceptual creativity, and intensive engagement with AI progress.

Conduct embedded evaluations of frontier AI labs and models to assess alignment and loss-of-control risks. Responsibilities include analyzing messy datasets, identifying anomalous model behavior, researching scheming and reward-seeking, developing evaluation tools and playbooks, writing high-stakes reports, and managing ambitious projects under tight deadlines. Some candidates will work on-site with frontier AI labs, while others will support evaluations through applied research and tooling.

Conduct empirical AI safety research focused on designing, implementing, and evaluating monitoring and control protocols for coding agents. Responsibilities include building evaluation frameworks and datasets, fine-tuning open-source models, developing agentic monitoring systems, analyzing failure modes, and optimizing detection accuracy, cost, and latency for production use.