ARBOx

Alignment Research Bootcamp Oxford

A 2-week intensive designed to rapidly build skills in AI safety.

ARBOx5: January 2027 (provisional)

Applications for the next ARBOx are not open yet.

Register interest ↗
Length
2 weeks, full-time
Where
Trajan House, Oxford
Cost
Free for admitted participants
Who
Open to all, not just Oxford students

Last time

ARBOx alumni have gone on to become MATS scholars, LASR participants, AI safety researchers and engineers at organisations like Anthropic, EleutherAI and Redwood Research, and to pursue PhDs at world-leading universities.

ARBOx4 is an in-person, full-time programme running from 28 June to 10 July 2026 at Trajan House in Oxford.

Programme Details

We provide Central Oxford accommodation for non-Oxford residents, lunch, and potential support with travel expenses. The programme is free for all admitted participants.

You do not need to be an Oxford student to participate (though we’d love to see a large number of applications from Oxford students!).

Technical Research Stream

During the programme, you’ll follow a compressed version of the ARENA syllabus, build gpt-2-small from scratch, learn interpretability techniques, understand RLHF, and replicate a paper or two. There will be lectures in the morning covering aspects of the syllabus, and afternoon pair-programming with TA support. During lunch break, there will be short talks from experts in the field, and we will host socials in the evenings. By the end, you’ll have hands-on experience with ML and safety techniques, a concrete mini project, and clear next steps for contributing to AI safety.

Syllabus

We’ll likely cover:

  1. 01Building gpt-2-small from scratch
  2. 02Foundational mech-interp techniques (e.g. steering vectors)
  3. 03Replicating some frontier alignment research (e.g. Emergent Misalignment)
  4. 04A brief introduction to RL and RLHF

Who should apply?

We’re looking for applicants with basic familiarity with linear algebra, Python programming, and AI safety (e.g., having completed AI Safety Fundamentals or another fellowship). Application is competitive, though: many more applicants meet the technical requirements than we have places.

We’d encourage you to apply if you’re eager to work on technical AI safety and satisfy the technical prerequisites. The programme is designed to upskill participants in ML safety, targeting those who would benefit from this training, regardless of their background. If you have relevant graduate or professional experience, this is also a plus! It’s easy to underestimate your abilities and your potential to contribute to the field, and so we suggest you err on the side of applying.

Generalist Stream

We are no longer planning to run the generalist stream. However, we may do so in the future. If you’d like to hear about if/when we open applications for future iterations of generalist ARBOx, please submit this form and we’ll be in touch.

If you would like to express interest in future iterations of ARBOx, feel free to do so here.

If you have someone in mind, feel free to mention them in our referral form.

Get involved

If you’re new and interested in learning more about Oxford’s AI safety community, we’d love to have you sign up for our mailing list or join us at an upcoming event.

If you have questions about AI safety or OAISI, you’re also welcome to reach out — we’re happy to meet for a coffee, go for a walk, or have a quick call.

Our newsletter.