Apollo Research is a non-profit organisation based in London that works on reducing catastrophic risks from advanced AI systems. The organisation brings together scientists, engineers, and policy experts to address AI safety challenges, with a particular focus on AI alignment and understanding systems that might covertly pursue misaligned objectives - a research area the organisation terms "scheming AI."
The organisation conducts pre-deployment evaluations of frontier AI models to detect potential risks before they reach users, and develops safety protocols for AI systems and deployments. This work involves close collaboration with leading AI research organisations including DeepMind, OpenAI, Anthropic, and Microsoft. The team has grown to approximately 35 members and recently expanded operations with a new office in San Francisco.
Apollo Research's work spans technical AI safety research, model evaluation methodologies, risk detection, alignment research, and AI policy. The organisation emphasises truth-seeking, goal-orientation, and constructive feedback as core elements of its approach to tackling some of the most complex challenges in AI development.
