London AI Safety Research (LASR) Labs

LASR Labs (London AI Safety Research Labs) is a technical AI safety research program operated by Arcadia Impact, a UK registered charity (charity number 1212323). The program evolved from AI Safety Hub Labs and was adopted by Arcadia Impact in 2024. It runs intensive 13-week cohorts, typically in summer and spring, where participants work full-time and in-person at the London Initiative for Safe AI. Participants are organized into teams of three to four and supervised by experienced AI safety researchers. Over the course of the program, each team takes a research project from initial proposal through to a completed academic-style paper and blog post. The program's philosophy is "learning by doing," and it specifically targets individuals planning to transition into technical AI safety careers within roughly a year. Research areas have included interpretability probes and concept extrapolation, AI control, deception in LLMs, multi-agent systems and collusion, scalable oversight, capability evaluations, specification gaming, and sparse autoencoders. Papers from the 2023 cohort were accepted to NeurIPS and ICLR workshops. All five papers from the 2024 cohort were accepted to NeurIPS workshops, and further papers from spring and summer 2025 cohorts achieved NeurIPS acceptance including an oral presentation. As of early 2026, LASR Labs has supported over 50 researchers and produced 14 research papers in technical AI safety. Participants receive an £11,000 stipend for living expenses in London, along with food, office space, and travel support. The program draws highly competitive applications (approximately 500 per cycle) and involves a multi-stage selection process. LASR Labs is staffed by three core team members: Erin Robertson (Programme Director), Brandon Riggs (Programme Manager), and Joe Hardie (Operations). The program is funded primarily through a grant from Coefficient Giving, which has provided over £8 million in total to Arcadia Impact since its founding.