Programs Directory

The real programs, grouped by where they sit on the ladder. This is a shortlist, not the whole world — the ones I’d actually point someone at if they asked me over coffee.

Read this before trusting any number below

Dates, stipends, and cohort timing change every single cycle, and several 2026 cohorts have already opened and closed as of today (June 2026). Treat everything here as “roughly what to expect,” then verify on the official page before you act. Honestly, learning to check primary sources is part of the training — never apply to anything based on a secondhand summary, including mine.

Quick scan

ProgramTrackLengthWhereBeginner-friendly?
BlueDot FundamentalsSafety / Policy2 hrs–8+ wksOnlineYes — start here
ARENAEng / Interp~5 wksLondonYes (some coding)
SPARSafety researchpart-timeRemoteYes
MATSSafety research10 wks +extBerkeley / LondonSome experience
LASR LabsSafety research13 wksLondonSome experience
PivotalSafety / Gov research~9 wksLondonSome experience
ERASafety / GovsummerCambridgeSome experience
Anthropic FellowsSafety research4 moSF / LondonStrong
AstraSafety research/eng~6 moBerkeleyStrong
GovAI SummerPolicy3 moLondonYes
HorizonPolicyup to 2 yrsUS (DC)Varies
RAND TASPPolicy~1–2 yrsUSSome experience
OpenAI ResidencyApplied / Research6 moSFStrong builders
DeepMind Student ResearcherResearch / Eng12–24 wksVariousEnrolled students

Beginner on-ramps (start here)

BlueDot - AI Safety / Governance Fundamentals

The standard front door. Free, online courses for people new to AI safety, no technical background required for the intro. Ranges from a ~2-hour “Future of AI” intro to 8+ week cohort courses (technical Alignment, or Governance) with weekly seminars.

My advice

Both of you should start the intro this week regardless of track. It’s a couple of hours and it gives us shared vocabulary for the rest of the sprint.

SPAR - Supervised Program for Alignment Research

A part-time, remote, mentored research program — you do real AI-safety research alongside other commitments (think ~5–40 hrs/week, flexible). Great first taste of supervised research without uprooting your life.

ARENA

The Alignment Research Engineer Accelerator. A ~5-week, in-person ML bootcamp in London built to turn someone with basic coding into a technical AI-safety Research Engineer. Covers deep-learning fundamentals, transformer interpretability, RL, and LLM evals. Travel, accommodation, and meals are covered.


Technical safety research fellowships

MATS

ML Alignment & Theory Scholars — the flagship mentored research fellowship. ~10 weeks of full-time research paired with a senior mentor, in Berkeley or London, with the possibility of a 6–12 month funded extension. Streams span interpretability, oversight, control, evals, governance, and more.

LASR Labs

A ~13-week, full-time, in-person research programme in London (based at LISA), producing a real research paper as a team with mentorship. Comes with a stipend (recent cohort ~£11k) plus office, meals, travel.

  • Track: Alignment and AI Safety research.
  • Who: some technical background; ready for a focused full-time research push.
  • Timing: summer cohort (recent one ran ~July–Oct).
  • Link: see the EA Forum announcement and LISA — search “LASR Labs.”

Pivotal Research Fellowship

A ~9-week, full-time research fellowship (London / LISA) across AI safety and governance, with a stipend (~£6–8k recent) plus housing, meals, travel, compute.

ERA Fellowship

A summer research fellowship based in Cambridge (UK), spanning technical safety and governance, with mentorship and a structured cohort.

  • Track: Alignment and AI Safety / Policy and Governance.
  • Who: students and early-career researchers.
  • Timing: summer.
  • Note: ERA also publishes a genuinely useful curated list of AI safety & governance opportunities — worth finding.

Anthropic Fellows Program

A ~4-month program where fellows do empirical safety research aligned with Anthropic’s priorities, aiming for a public output (like a paper), with mentorship from Anthropic researchers. Recent cohorts cover scalable oversight, adversarial robustness & AI control, model organisms, mechanistic interpretability, AI security, and model welfare. Well-funded (recent cycle: weekly stipend plus substantial compute).

This is the one I did

I went through this program, so take the rest of this list with that bias in mind. What I’d tell anyone applying: the bar is a real, legible piece of technical work plus a specific reason this research area pulls at you — not polish, not buzzwords. The application itself taught me more about my own direction than the month of prep before it.

>

Astra Fellowship

Run by Constellation (Berkeley). A fully-funded, in-person program pairing fellows with senior mentors for research/engineering on AI safety, ~6 months. Strong placement record — a large share of early fellows went on to full-time safety roles at major labs and institutes.


Policy & governance fellowships

GovAI Summer Fellowship

The Centre for the Governance of AI’s flagship. A 3-month fellowship in London to launch a career in AI governance, with two tracks: Research (independent project with a supervisor) and Applied (non-research — comms, advocacy, events, research/program management, ops, fundraising). No specific degree required; the process includes a written submission, an automated assessment, a paid remote work test, and an interview. Stipend (recent applied track ~£12k).

Horizon Fellowship

Horizon Institute for Public Service places fellows into US government and think-tank roles working on emerging tech (including AI) policy. Longer commitment (up to ~2 years), with training and placement support. The serious route into Washington AI policy.

RAND Technology and Security Policy Fellowship

A fellowship at RAND (a major policy research institution) focused on technology & security, including AI. Rigorous policy-research environment; strong for the analytical/research side of governance.

  • Track: Policy and Governance, research.
  • Who: strong analytical/writing background.
  • Link: search “RAND Technology and Security Policy Fellows.”

Also worth knowing (US policy)

TechCongress places technologists into US congressional offices — more of a tech-to-policy bridge than an AI-specific program, but a real route if either fellow is US-based and policy-curious.


Industry research & residencies

OpenAI Residency

A 6-month program for talented people from adjacent fields (math, physics, neuroscience, strong self-taught engineers) to move into frontier AI research/engineering. Residents are paid as full employees and can convert to full-time.

Google DeepMind Student Researcher

Not a “fellowship” but the best internship-style research on-ramp if you’re currently enrolled. 12–24 weeks, embedded with DeepMind researchers. DeepMind also runs PhD scholarships, postdoc fellowships, and the undergrad-focused Research Ready programme (UK).

Also in this category


How to use this list

Pick a primary target and a backup, and ideally line up at least one beginner on-ramp to run in parallel. Then point everything — your skills, your project, your application — at those. Don’t apply to twelve things; apply well to two or three that actually fit. (Applications (the actual mechanics))


Back to Home · Tracks Overview · Skills Map