Mercial.

How can we mitigate catastrophic risk from advanced AI?

The Mercial AI Safety Initiative is a community working on the technical and strategic problems of making powerful AI systems safe. We run a remote research fellowship, hackathons, and a hands-on upskilling bootcamp. All of them are open to anyone.

Apply to the research fellowship See all our programmes

What we do

Research fellowship

Fellows work on a technical or conceptual problem in AI safety with a facilitator over a month-long remote fellowship, and produce a write-up, building real research experience. Applications for the November cohort have closed, apart from two streams open until 1 November.

Hackathons

Intensive weekend sprints, including the Apart Research hackathons — a fast way to try a research idea and see it through to a write-up. Upcoming dates.

Technical bootcamp

An intensive upskilling programme built on the ARENA curriculum: transformer internals, mechanistic interpretability, RL and evals. Read more.

Why AI safety

AI capabilities are advancing quickly, and the field of people working to make these systems safe, interpretable and controllable is still small. Applicants with strong conceptual, mathematical or technical backgrounds are unusually well placed to contribute.

You don't necessarily need prior machine learning experience to get involved. See Get Involved for how you can gain the background knowledge you need for AI safety research.