Termite is an interdisciplinary R&D studio for real-world AI alignment, based in London.

We run intensive programmes that bring together practitioners from relevant disciplines to anticipate the problems that arise when AI enters real-world technical, economic and social systems, and develop the interventions needed to preempt them. Of particular concern are the ways in which systems and contextual factors beyond the sandbox can influence the behaviour of agentic AI, leading in turn to the distortion of how these systems function. We call this “alignment in the wild”.

These systems include:

• Financial infrastructure
• Information environments
• Scientific research
• Labour markets
• Corporate and civic governance

We believe that the complexity of AI alignment as an area of study cannot adequately be addressed by technical expertise alone. It also requires the skills and perspectives of those working across the natural and social sciences, law and policy, the humanities and the arts. Researchers from these disciplines can identify aspects of AI alignment that purely technical practitioners miss, improving how experiments are designed, their outcomes are interpreted and solutions are built.

Each Termite programme convenes a small cohort of multi-disciplinary researchers at our studio space in London, where they are given the resources and structure needed to interrogate specific areas of real-world alignment through targeted experiments and empirically grounded speculation.

Programme outputs are principally written for those building and regulating affected real-world systems, rather than developers of the AI models themselves. However, we release all of our research and methods publicly.