← Back to all writing

Personal notes: resources for getting started in AI safety

February 1, 2026

When I wrote about who is governing AI and the happy path, one question kept coming up: what can an individual actually do?

This is my personally curated starter list — not an exhaustive catalog, but sorted from “do today” to “years of preparation” so newcomers can find an entry point. For the full visual index, see the AISafety.com Field Map (300+ orgs, filterable by advocacy / research / training / funding, etc.) — this post is my filtered on-ramp; the map is the panorama.

I know there are plenty of opportunities in China, the UK, Singapore, and elsewhere — but I’m based in the Bay Area, so what follows is inevitably biased toward what’s around me. Take it selectively.

Dozens of organizations do serious work on tiny budgets — $200-400 million globally for AI safety versus $252 billion in corporate AI investment. Many of them need people more than they need ideas.

How to read the ratings:

  • ⭐ Anyone (interest is enough)
  • ⭐⭐ Some background (undergrad degree, technical skills, or relevant experience)
  • ⭐⭐⭐ Professional background (graduate degree, research experience, or industry experience)
  • ⭐⭐⭐⭐ Expert (PhD + publications, professor, senior industry)

The real on-ramps: community, career planning, and programs

Most people who enter AI safety don’t start by applying to a research org directly. They enter the EA / LessWrong community ecosystem first — career advising, training programs, funding channels, and social graph. This infrastructure layer matters as much as any org below.

Career planning and information hubs

80,000 Hours — The most systematic AI safety career guide; free 1-on-1 advising; AI safety career guide and podcast are community standards. If you’re still figuring out what to do, start here.

LessWrong / Alignment Forum / EA Forum — Where alignment research and governance discourse happen first — often before journals. Reading The Sequences is the baseline for understanding how this community thinks. ⭐ Read / ⭐⭐ Discuss or post

Effective Altruism community — AI safety is a core focus. Local groups in most major cities; EAG / EAGx are the highest-density networking events. ⭐

ACX Meetups Everywhere / First Saturday — Lowest-barrier offline entry to the rationalist community — more casual than EAG. ⭐

Physical hubs: Lighthaven (Berkeley), Constellation, Mox (SF) — LessOnline, Manifest (forecasting conference), ACX monthly meetups. Showing up beats reading a hundred reports.

Jobs and signal sources

ResourceWhat it’s for
80,000 Hours Job BoardPrimary aggregator for AI safety / governance roles — Redwood, METR, GovAI, frontier lab policy, etc.
AISafety.com Field MapCommunity-maintained panorama — 300+ orgs/programs, browse by category; also links to jobs, funding, events, volunteer projects
AISafety.comResource hub homepage (courses, communities, advisors, etc.)
Org careers pages directlyAnthropic, OpenAI, Redwood, METR, Apollo, UK AISI, etc. — 80k isn’t complete; big labs often post only on their own sites
Rohin Shah career FAQRole taxonomy, PhD or not, inside-view research methodology — more field-specific than generic career advice
Newsletters to scanCAIS AI Safety Newsletter, Anthropic / Alignment blog, GovAI, Stanford HAI, Import AI (Jack Clark), Alignment Newsletter archive (Rohin Shah — stopped but canonical)
Slack / DiscordAI Alignment Slack (apply to join), EleutherAI Discord (open-source ML / interp crowd)

Training programs and fellowships

ProgramDuration / formatWho it’s forBarrier
MATS (ML Alignment & Theory Scholars)10–12 weeks, Berkeley or London; stipend + compute + housingTechnical background, want mentored research with frontier mentors. One of the community’s core on-ramps.⭐⭐⭐ Competitive (~5% acceptance)
Anthropic Fellows4 months full-time, paid + computeTargeting frontier lab alignment research⭐⭐⭐⭐ Rolling applications
AI Safety CampMulti-week, remote, freeEngineering/ML background but not in alignment yet. One of the friendliest on-ramps for newcomers.⭐⭐
SPAR (Supervised Program for Alignment Research)3 months, remote part-time, no stipendWant to try mentored research without going full-time⭐⭐ One of the lowest barriers
BlueDot Impact (AI Safety Fundamentals)12 weeks ~5 hr/week, free cohortZero background, want an entry credential
Astra Fellowship (Constellation)5 months in-person, BerkeleyFull-time empirical or governance research⭐⭐⭐⭐
ARENA4–5 week London bootcampNeed to quickly build Python/ML foundations⭐⭐
OpenAI Residency6 months full-time, high stipendEngineering/builder background into OpenAI, no PhD required⭐⭐⭐⭐ Annual cohort
Horizon Fellowship6–24 months, DCFederal agencies / Congress / think tanks for AI policy⭐⭐⭐ Rolling
Pivotal Research Fellowship9 weeks, LondonPipeline to GovAI / DeepMind / PhD⭐⭐⭐
CAIS / GovAI / FLI fellowshipsVariesPolicy/governance or research tracks⭐⭐–⭐⭐⭐

Typical pipeline: BlueDot or AI Safety Camp → MATS or SPAR → Anthropic Fellows / lab full-time. You don’t need a PhD to do serious work — many alignment researchers take the fellowship pipeline, not the traditional academic track.

Funding and independent research

SourceUseBarrier
LTFF (Long-Term Future Fund)Independent research, typically $30–80k/yr⭐⭐
ManifundCommunity grants, ACX Grants, fiscal sponsorship⭐⭐
Coefficient Giving (formerly Open Philanthropy)Career transition funding, technical safety RFPs⭐⭐⭐
FLI grantsResearch grants (~$17M annual budget)⭐⭐⭐
SFF (Survival and Flourishing Fund)Mid-size project regranting⭐⭐

Want 6–12 months off to focus on research? LTFF or a Coefficient career grant is the standard EA-community path for independent research — you don’t need to join an org first.

Field building and startup paths

Research isn’t the only entry. If you want to build orgs, products, or talent infrastructure rather than pure alignment research:

ResourceWho it’s forBarrier
Halcyon FuturesFounders/operators doing 0→1 safety companies or nonprofits. Career transition grants (up to ~$250k), org seed grants (up to ~$500k), co-founder matching; paired with Halcyon Ventures. Also mentors the MATS Founding & Field-Building stream (Mike McCormick, Nick Raushenbush)⭐⭐⭐ Need something you’ve built
MATS Founding & Field-Building track10 weeks in Berkeley/London on a product/org or talent-matching project⭐⭐⭐
ManifundLaunch small projects fast (fiscal sponsor, ACX Grants)⭐⭐
FAR.AIResearch incubator; also spins out orgs (e.g. Safe AI Forum)⭐⭐⭐
Apart ResearchHackathon → Studio → Fellowship — rapid safety tool/project prototypes⭐⭐

Halcyon’s Request for Founders lists field-building as its own sector — talent pipelines, recruiting infra, public comms all count.

Volunteer opportunities

Two buckets: advocacy/watchdog (public engagement) and field-building (community infrastructure).

OrgWhatBarrier
The Midas ProjectAI company watchdog — research, scorecards, transparency/accountability advocacy; official volunteer signup
Pause AILocal chapters, PauseCon DC lobbying events
Encode AIYouth advocacy org (SB 53, RAISE, etc.); local chapters⭐ Skews student/youth
FLIOpen-letter campaigns, event support
Center for Humane TechnologyOutreach, volunteer projects⭐–⭐⭐
Alignment Ecosystem Development (AED) / AISafety.comMaintains the Field Map, community directory, funding newsletter, ~30+ volunteer projects (web dev, SEO, user research, data); Discord with monthly onboarding calls⭐–⭐⭐ Eng/product background helps
AISafety.info (Rob Miles / Stampy)Maintain FAQ and chatbot knowledge base⭐⭐
AI Safety QuestNewcomer navigation, coaching, outreach
AI4ALLMentor high school AI programs
DataKindData/ML projects for nonprofits⭐⭐ Needs coding
LessOnline / EAGConference volunteering
Lightcone InfrastructureRuns LessWrong, Alignment Forum, and the Lighthaven campus — core infra for the rationalist/alignment ecosystem. LessOnline / Manifest event volunteers; Berkeley Sequences Reading Group (Tuesdays); day-to-day event/campus support via habryka@lighthaven.space or LW Intercom⭐ Easier if you’re in the Bay Area
Local EA groupsHelp run monthly meetups

How to pick: Research track → AED/AISafety.com or AISafety.info beats Pause AI for career graph. Policy/advocacy → Midas, Encode, Pause AI. Eng/product → AED’s projects page has many roles seeking owners.

Volunteering usually isn’t the highest leverage, but for newcomers still finding direction, it’s low-cost trial and error plus meeting people in the space.


Things anyone can do today

Before the organization list — these take minutes and require no credentials.

ActionWhereWhat it takes
Browse the AISafety.com Field Map or 80,000 Hours Job BoardOnline10 minutes
Use 80,000 Hours for career planning (or book a free advising call)Online30 minutes
Register on LessWrong / Alignment Forum and start readingOnlineReading ability
Join a local EA group or attend EAGxMost cities globallyShow up
Sign the AI safety open letterCBPAI / FLI / CAISAn email address
Write to your representative supporting AI regulationYour district5 minutes. 70-86% of the public supports AI regulation — politicians need to hear it
Read one report from METR, GovAI, or Stanford HAIOnline30 minutes
Attend a Pause AI eventGlobal chaptersShow up
Share reliable AI information (not hype) on social mediaYour accountsJudgment
Donate to an underfunded AI safety orgFLI, CAIS, METRAny amount

These actions look small, but they’re low-cost for newcomers. Every person who contacts a representative, joins a group, or donates shifts the equation at the margin.


United States

Research organizations

MIRI (Machine Intelligence Research Institute) — The oldest alignment research organization (founded 2000). Theoretical AI alignment research focused on mathematical foundations. Berkeley, CA.

  • ⭐⭐⭐⭐ Research positions require strong math/CS. Anyone can read their work and support through donations.

CHAI (Center for Human-Compatible AI) — Stuart Russell’s alignment research center at UC Berkeley. Focuses on the problem of AI systems that are uncertain about human preferences.

  • ⭐⭐⭐ Research positions need PhD-level. Sometimes hires undergrad research assistants. Open lectures and seminars.

METR (Model Evaluation & Threat Research) — Measures what AI can actually do. Produced the task horizon data showing AI capability doubling every 3-7 months. Their evaluations are what governance depends on.

  • ⭐⭐⭐ Research and engineering roles need strong technical background. Their reports are public and worth reading for anyone.

ARC Evals (Alignment Research Center) — AI safety evaluation frameworks. Founded by Paul Christiano (former OpenAI alignment lead).

  • ⭐⭐⭐ Research positions need ML/safety background.

Apollo Research — Detects deceptive AI behavior. Found that 5 of 6 frontier models exhibit “scheming” — covertly pursuing misaligned goals.

  • ⭐⭐⭐ ML research background. London / remote.

Redwood Research — Technical alignment research (interpretability, adversarial training). Very selective. Berkeley, CA.

  • ⭐⭐⭐⭐

FAR.AI — AI safety research incubator. Helps early researchers start alignment projects. Lower bar than MIRI/Redwood — explicitly designed as an on-ramp.

  • ⭐⭐⭐ Paid fellowships. Berkeley, CA.

Policy and governance

Center for AI Safety (CAIS) — Research and advocacy. Organized the “AI extinction risk” statement signed by Hinton, Bengio, Altman, and hundreds of researchers. Has fellowship programs for grad students and early-career researchers.

  • ⭐⭐ Fellowship open to graduate students. San Francisco.

AI Now Institute — Social impact of AI: labor, fairness, corporate power. Based at NYU.

  • ⭐⭐ Research positions need social science/policy background. Public reports available.

AI Policy Institute — Direct policy advocacy and lobbying for AI safety legislation. Washington DC.

  • ⭐⭐ Policy or communications background.

OpenResearch — Open-ended social research. Ran the largest US UBI experiment (backed by Sam Altman). San Francisco.

  • ⭐⭐⭐ Social science / economics background.

AI Futures Project — AI forecasting and scenario planning. Founded by Daniel Kokotajlo (former OpenAI governance researcher). Produced the AI 2027 scenario. Berkeley, CA.

  • ⭐⭐⭐ AI / policy background.

Academic centers

Stanford HAI — Publishes the annual AI Index (the most cited quantitative overview of AI). Research, policy analysis, and public events. Has undergraduate research opportunities.

  • ⭐⭐⭐ for research. ⭐ for public events and reports.

MIT FutureTech — Technology’s impact on work and society.

  • ⭐⭐⭐

Advocacy

Future of Life Institute (FLI) — Organized the 2023 pause letter (33,000+ signatures). Also distributes research grants ($17M annual budget). Boston / global.

  • ⭐ Anyone can sign letters, attend events, donate. ⭐⭐⭐ Apply for research grants.

Pause AI — Grassroots movement organizing public protests and advocacy for pausing frontier AI training. Has chapters globally.

  • ⭐ Anyone can attend events and join. Volunteer-run.

Center for Humane Technology — Technology ethics advocacy (Tristan Harris, Aza Raskin). Produced “The AI Dilemma” and other public talks.

  • ⭐ Public content available to all. ⭐⭐ Volunteer opportunities.

Public benefit / AI for good

Partnership on AI — Multi-stakeholder AI governance dialogue. Members include companies, academics, and civil society.

  • ⭐⭐ Organizations can apply for membership.

AI4ALL — Increasing diversity in AI. Programs for high school students. Adults can mentor or volunteer.

  • ⭐ High schoolers can apply. Adults can mentor.

DataKind — Data science for social good. Volunteer-driven projects helping nonprofits.

  • ⭐⭐ Needs data/programming skills. Volunteer.

International

GovAI (Centre for the Governance of AI, Oxford) — Arguably the most influential AI governance research organization. Published the compute governance framework and international coordination proposals. Has a competitive summer fellowship for PhD students.

  • ⭐⭐⭐ Research fellowship needs strong academic background. Their research reports are public and extremely high quality.

Ada Lovelace Institute — Social impact of AI and data. UK and European policy focus. London.

  • ⭐⭐ Research positions and public engagement.

Cambridge LCFI (Leverhulme Centre for the Future of Intelligence) — Long-term AI impact: philosophy, ethics, society. Cambridge, UK.

  • ⭐⭐⭐ Research positions. Public seminars.

Community ecosystem, training programs, fellowships, and funding — see The real on-ramps above. That section matters more than the org list below.


China

The AI safety and governance community is much thinner in China. Most organizations are government-affiliated research institutes rather than independent civil society.

CAICT (中国信通院) — Ministry of Industry research institute. Publishes AI white papers, participates in policy design.

  • ⭐⭐⭐ Internal researchers. Public reports available.

Tsinghua AI Research Institute (清华大学人工智能研究院) — AI research and governance. Led by Academician Zhang Bo.

  • ⭐⭐⭐⭐ PhD level.

BAAI (北京智源人工智能研究院) — Published the “Beijing AI Principles.” Some alignment-adjacent work.

  • ⭐⭐⭐

Concordia AI — One of the few organizations bridging US-China AI safety dialogue. Bilingual research and exchange. If you’re bilingual and care about this space, this is uniquely valuable.

  • ⭐⭐⭐ AI / policy background + English-Chinese bilingual.

Chinese EA community (有效利他主义中国) — Small but growing. Local events in major cities.

  • ⭐ Interest + showing up.

For general awareness in Chinese: 机器之心, 量子位, AI前线 (WeChat public accounts) cover AI developments. For AI safety specifically, resources are sparse — most serious discussion happens in English, which is itself a barrier.


Optional: governance engagement

Not the main path — technical readers should prioritize the MATS/SPAR pipeline above; non-technical readers don’t need to become political commentators. But if you care about specific legislation rather than abstract x-risk, you can engage in a targeted way:

NYC example: Alex Bores (State Assembly AD-73, co-author of the RAISE Act) is a concrete model of state-level frontier AI legislation — read his AI policy framework, follow RAISE implementation and amendments in Albany, contact your NY Assemblymember about AI bills. He lost the NY-12 congressional primary in 2026 but remains in Albany on state-level work — state law often moves faster than federal.

California example: Scott Wiener’s SB 53 (frontier AI transparency/safety framework) is the parallel front line.

Policy career track: For full-time policy work, see Horizon Fellowship (DC), GovAI, AI Policy Institute in the fellowship table above. That’s a different line from grassroots engagement.


What’s missing from this list

A few observations:

China’s civil society gap is the most striking. Almost all independent AI safety/governance organizations are in the US and UK. China has government-affiliated research institutes but almost no independent civil society working on AI safety. Given that China is one of the two most important countries for AI governance, this is a critical gap. Concordia AI is one of the only bridges.

Most organizations are severely underfunded. The entire global AI safety NGO sector operates on roughly $50-100 million per year — less than what it costs to train a single frontier model. Donations, even small ones, are disproportionately impactful.

The typical technical on-ramp is the EA/LW community + training pipeline: BlueDot or AI Safety Camp → MATS or SPAR → lab fellowship.

The typical field-builder on-ramp: Manifund/Apart small project → MATS F&FB track or Halcyon → full-time org/product work.

80,000 Hours, LessWrong, and EA aren’t side notes — they’re field infrastructure.

For non-technical readers: optional governance engagement (previous section) — contacting representatives, supporting specific bills, sharing evidence-based content — beats vaguely “following politics.” For technical readers: comparative advantage is research and engineering; governance literacy is useful background, usually not the main path.


About this post

This is my personally curated starter list — I’m a complete beginner too (left my job in 2026 to work on AI safety full-time; still figuring out MATS, fellowships, and research paths). If you’re just starting out and want to chat about directions, feel free to email me. I won’t have all the answers, but I can share what I’ve found so far.


Sources