MARS V

How is MARS different from other programmes?

MARS is part-time and hybrid. Unlike fully in-person, full-time programmes (MATS, ERA) or fully remote ones (SPAR, FIG), MARS is compatible with concurrent studies or employment for both mentors and mentees.

What research areas does MARS cover?

Technical AI safety (e.g., AI control, interpretability, evals, CoT faithfulness) and AI governance & policy (e.g., international agreements, compute governance, hardware-enabled mechanisms). Specific MARS V projects are still forthcoming.

When does the programme run?

MARS V’s in-person component will run during the Northern Hemisphere summer (June and/or July). Applications open late March with a mid-April deadline. If you can’t participate in MARS V but want to join MARS VI (December 2026), please fill out the mentee EOI above.

Is there any cost to participate?

No. MARS is free — CAISH provides travel funding, accommodation, office space, and meals during the in-person sprint week.

What support and resources are provided?

Teams receive a $2k+ compute budget, Claude Max (5x) for technical streams, a dedicated research manager, office space and catering during the sprint week, and travel funding and accommodation.

Who runs MARS?

The Cambridge AI Safety Hub (CAISH). Gaurav Yadav and Justin Dollman currently co-direct CAISH and run MARS. MARS was founded in 2024 by Chloe Li, former CAISH director and current Anthropic Fellow.

Time Commitment

Participants should expect to spend 8–15+ hours/week on their project. The programme includes 2–4 weeks of pre-work, a 1-week in-person kick-off in Cambridge, followed by 10–12 weeks of remote part-time research.

Weekly Check-Ins

Throughout the duration of the program, you are expected to check-in:

• With your teammates once a week. It’ll be up to your team to coordinate meeting times or create co-working systems.
• With your supervisor once a week. The format of check-ins with your supervisor will be at their discretion.

You will also be assigned a research manager who will support your work throughout the programme. While you’ll coordinate with your team, the research manager provides external accountability and guidance to help keep your project on track.

We will also have program-wide events and check-ins, including socials, co-working sessions, and guest speakers. More details on these events will be announced to accepted participants.

MARS Research Symposium

This event will take place at the end of the program, where each research group presents their research.

Who should apply to mentor?

Researchers with experience in AI safety who have a concrete project idea and want to supervise 2–4 early-career researchers. We’ve had academics (professors to late-stage PhD students), research engineers, and professionals from Redwood Research, UK AISI, MIRI, Geodesic Research, GovAI, DeepMind, and more.

We’re decidedly non-credentialist. As long as you have a well-scoped project with a theory of impact, we will take your application seriously.

What will I need to provide?

• A project description (theory of change, what mentees will do, expected output)
• Application questions for prospective mentees (coding tests, essays, research design tasks — whatever fits your project)
• Regular mentorship time during the remote phase (weekly team meetings at minimum)
• Attendance at the in-person kick-off week is strongly preferred but not absolutely required

Time commitment

In-person (1 week): Work alongside your mentees in Cambridge. Remote (~12–16 weeks, part-time): Expect 1–3 hours per week for group meetings and async communication. The exact structure is up to you.

Who will I mentor?

University students and early-career researchers with an interest in AI safety. Most mentees have completed introductory AI safety programmes and have some coding/research experience. You’ll typically have 2–4 mentees on your team, and you run Stage II of the application process, giving you complete control over your team’s composition.

What support do I get?

• A curated applicant pool (saving you time during selection)
• A research manager who handles logistics, runs weekly check-ins with your team, and flags problems early
• Office space and infrastructure during the in-person week
• Compute budget for your team if needed

What kind of output is expected?

Almost without exception, the deliverable is one or more publications — LessWrong posts, conference papers, arXiv preprints, or academic articles.