Research Scientist/Engineer (Fundamental Research)

The role

This person will do fundamental AI safety research, studying how reinforcement learning leads to reward-seeking, evaluation awareness and misaligned preferences. They will design evaluation techniques and build a "Science of Scheming" in collaboration with leading AI developers.

What you will do

• Collaborate with leading AI developers through long-term research partnerships.
• Study the RL dynamics behind reward-seeking, evaluation awareness and misaligned preferences.
• Design and train model organisms and scale insights to frontier systems.
• Build empirical foundations to predict how scheming risks evolve as models scale.
• Develop novel evaluation techniques that could scale to highly evaluation-aware models.
• Investigate patterns in the reasoning processes of frontier AI systems.

Required qualifications

• Design and execute experiments, driving fast empirical iteration cycles.
• Use AI tools to accelerate your work and know which models suit which tasks.
• Familiarity with prior work on AI scheming, including classic and modern literature.
• Produce high-quality scientific writing, figures, blog posts and tweet threads.
• Stay current on model releases and tinker with AI workflows.
• Hands-on experience training LLMs via reinforcement learning.
• Rigorous quantitative skills, such as from scaling laws, statistical physics, dynamical systems or applied statistics.
• Commitment to the mission and to an intense, high-expectation culture.

Nice to have

• Strong software engineering skills; the stack is entirely Python.
• Project management experience, including managing strong individual contributors.

Benefits

• Market-competitive salary and equity.
• Unlimited vacation.
• Unlimited sick leave.
• Up to 6 months of paid parental leave.
• Comprehensive health, dental and vision insurance.
• Retirement savings with employer matching (e.g. 401(k) for US employees).
• Lunch, dinner and snacks provided on workdays.
• Paid work trips, including staff retreats, business trips and conferences.
• A yearly $1,000 professional development budget.
• Relocation support and visa fees where applicable.

Before you apply

• Visa sponsorship is offered in the UK and US, but not guaranteed for every role or candidate.
• The interview process has a screening interview, 3 technical interviews and a final interview with the CEO.
• Technical interviews relate to on-the-job tasks; there are no general LeetCode-style coding interviews.
• Candidates are encouraged to prepare by replicating or extending a recent paper or blog post, or by writing a project plan addressing a conceptual limitation.

About the company

Apollo Research works on risks from loss of control, particularly deceptive alignment and scheming, through detection, the science of scheming and mitigations. It works closely with multiple frontier AI companies, including testing their models before deployment.