BlueDot Impact

BlueDot Impact is a London-based non-profit talent accelerator that runs the world's largest structured courses on AI safety and governance, and whose curricula underpin the programmes of more than 100 local groups worldwide. Founded in 2022 out of informal reading groups at the University of Cambridge, it has since trained over 8,000 people — from ML researchers and software engineers to policy professionals and government officials — and supports the pipeline of talent entering AI-safety and AI-governance careers globally. The organisation is the primary curriculum infrastructure for the EA-adjacent AI-safety field-building layer; in this corpus, it is the closest analogue to a parent organisation for local hubs including BAISH in Buenos Aires and impactRIO in Rio de Janeiro, and a programmatic partner for Berlin AI Safety.

Founding

The organisation traces directly to 2021, when Jamie Bernardi — then at Cambridge — found himself unable to locate coherent resources for entering AI-safety work and began running reading groups that evolved into structured courses. Those early sessions became the AI Safety Fundamentals course, initially delivered informally from 2021. In August 2022, Bernardi and Dewi Erwan Humphreys — formerly Executive Director of Effective Altruism Cambridge and Biosecurity Advisor to the Cambridge Existential Risks Initiative — co-founded BlueDot Impact as a nonprofit startup in Cambridge, alongside William Saunter. The inaugural 2022 cohort enrolled 400 participants.

The organisation was formally incorporated in the UK on 27 June 2023 as BLUEDOT IMPACT LTD (company number 14964572), a private company limited by guarantee — the standard UK legal structure for non-profits without share capital. The company is headquartered at 71–75 Shelton Street, Covent Garden, London. Erwan Humphreys serves as CEO and director; Bernardi was a director from incorporation until early 2024, when he left to pursue AI policy work after a GovAI Winter Fellowship. Connor Campbell joined as a director in July 2025. The organisation expanded to San Francisco in 2025.

Courses

BlueDot's course portfolio is organised around a three-level structure designed to move learners from first contact with AI-safety ideas through to advanced specialisation.

Technical AI Safety. The centrepiece course, successor to the original AI Safety Fundamentals alignment curriculum. It targets ML researchers and software engineers and is delivered in a part-time cohort format — roughly five hours per week over two to three months, combining weekly synchronous facilitator-led sessions with asynchronous reading and exercises. The format is adapted from the original reading-group pedagogy: small cohorts, peer discussion, and a shared curriculum. It is free to participants.

AI Governance. A parallel cohort course for policy professionals, researchers, and practitioners working on AI regulation and governance questions. The governance track was introduced in 2023 alongside the redesigned alignment course.

Future of AI. A two-hour self-paced introductory course on AI capabilities and trajectory, open to all without application. It serves as the entry point for participants who are new to the field.

Biosecurity Fundamentals. A course on global catastrophic biological risks, piloted with 100 participants in 2023 and developed alongside the core AI-safety offering.

The underlying framework BlueDot uses to organise these tracks is a "defence-in-depth" model with five layers — Prevent, Detect, Constrain, Withstand, and Adapt — which it treats as parallel technical and governance challenges, rather than assuming any single approach to AI risk will be sufficient.

Local facilitator network

A distinctive feature of BlueDot's model is the local facilitator programme: more than 100 groups worldwide independently run BlueDot's curricula through their own in-person cohort courses. This model allows BlueDot's courses to be delivered in local languages, embedded in local academic and professional communities, and adapted to regional context — while using a shared, maintained curriculum.

In this corpus, BAISH (Buenos Aires AI Safety Hub) runs the Technical AI Safety course in Spanish at the Facultad de Exactas y Naturales, UBA; Eitan Sprejer, BAISH's founding director, serves as an official BlueDot course facilitator. impactRIO ran all three of its fellowship cohorts — Alignment (2023), Advanced Alignment (2024), and Governance (2024) — on BlueDot Impact curricula, contributing to 26 fellowship completions in its first year. Berlin AI Safety co-produced the AI Futures Forum on 21 June 2025 with BlueDot Impact, a day-conference positioning Germany's AI policy environment as the theme. These three entries represent different points on the spectrum of BlueDot engagement: BAISH and impactRIO use BlueDot's curriculum as their primary programmatic infrastructure; Berlin AI Safety is a peer event partner.

Scale and outcomes

By mid-2026, BlueDot reports 8,000+ alumni trained since 2022, drawn from backgrounds spanning ML research, software engineering, policy, government, and humanitarian organisations. Approximately 1,000 graduates now work in high-impact roles at organisations including Anthropic, Google DeepMind, the UK AI Security Institute, the United Nations, NATO, Stanford HAI, and the Harvard Kennedy School. The organisation raised $35 million in total — $5 million in 2024 and $25 million in 2025 — enabling expansion to monthly course runs and a San Francisco office. Both tranches came from Coefficient Giving (formerly Open Philanthropy) as general support grants.

The 2024 growth trajectory reflects the field-building dynamic: from 400 participants in its inaugural 2022 cohort to 3,500 trained in 2024 alone. The shift to monthly courses in 2025 further compressed the cycle between cohort launches.

Place in the movement

BlueDot Impact occupies an unusual structural position in the make-AI-good movement's grassroots layer. It is not a campaign organisation, a policy advocacy group, or a research institute; its product is the trained talent that fills those roles, and the local group infrastructure through which that training is delivered at scale. The 100+ local facilitator groups are the grassroots face of BlueDot's reach — the human scale at which students in Buenos Aires, Rio de Janeiro, Berlin, and elsewhere are introduced to AI-safety ideas and connected to the research and policy communities that form around them.

The scope question this entry surfaces — "course-based field-building vs grassroots-engagement test" — resolves on the evidence. BlueDot's courses deliberately target people not yet working on AI safety (ML engineers, policy professionals, students, government officials) and move them into the field; the 100+ local group network is the mechanism through which that movement of people happens at community scale. This places BlueDot in the same "community-building substrate" category the 2026-05-23 ruling codified for local AI-safety groups — the difference is that BlueDot operates one layer above, as the infrastructure those groups depend on rather than as a local group itself.