Made with ♥ in Silicon Valley. Copyright © Joy Larkin 2002-2026.
contact
•••
speaking
•••
cleverhack.com
•••
llms.txt
•••
cleverhack.com
Frontier AI Lab News Tracker ⇒
Neolab and Emerging AI Lab Tracker ⇒
Who Tests The Frontier AI Labs? What We Know ⇒
Have news to share or need to update/add info?
Email: cleverhack at duck.com
LinkedIn: /in/joylarkin
Twitter: @joy
November 4, 2026 2:45 PM PT
Most writing about getting into frontier AI assumes you are already in the pipeline — a PhD student, a lab intern, someone with a referral. This page is for everyone else.
These are the structured entry points into frontier AI and AI safety research: programs that take applications from outside, pay a stipend or cover costs, and end with you having done real work. Several were designed specifically for career-changers, and most say plainly that no PhD and no prior ML publications are required.
Considerations for inclusion:
- Open to applicants from outside the lab and academic hiring pipelines
- A defined cohort, curriculum, or mentorship structure — not just a job posting
- Funded, or costs covered, or free to the participant
- Currently operating, with a named upcoming or recent cohort
A note on timing. As of October 2026, many cohorts for the coming winter have already closed. That is normal and it is the single most useful thing to understand about the ecosystem: programs run on annual or semi-annual cycles, and the application windows are short. Each entry below lists when applications typically open so the page stays useful between cycles. Nearly every program keeps an expression-of-interest form — so filling those out is the highest-value thing to do when nothing is open.
Recently added as of November 4: the Anthropic Residency, a paid research placement of six weeks to six months for established technical leaders, in Research Programs; and two programs that came out of the Generator Residency — M3, a paid fellowship for senior security professionals, in Career-Changers, and the Longitude DC Intensive for technical people moving into policy, in Policy & Governance. Also new since October 20: the HIIG Impact AI Fellowship (closes January 15) and CSR’s AI and National Security Convergence Fellowship in Policy & Governance, Current’s AI Futures Mastermind in Career-Changers, a new AI Security Bootcamp cohort in San Francisco, Apart’s Swarm Alignment Hackathon (November 6–8), and ARC’s White-Box Estimation Challenge.
Of interest: GovAI’s Summer 2027 fellowships are open until November 22, including a London Applied Track for communications, policy, operations, and program management professionals.
Deadlines to watch:
➟ November 7 — AI Security Bootcamp, San Francisco, January 2027
➟ November 8 — BlueDot AGI Strategy course
➟ November 11 — BlueDot Incubator Week, San Francisco
➟ November 15 — BlueDot Frontier AI Governance course
➟ November 16 — Iliad Fellowship and Iliad Intensive, January 2027
➟ November 22 — Anthropic Residency; Tarbell Fellowship 2027–28; GovAI Summer Fellowship 2027, London and Washington, DC
➟ November 29 — AI Safety Camp team applications (open November 13)
➟ December 16 — Iliad Fellowship, February 2027
Research Programs | Research Organizations | Evaluators & Red-Teaming | Policy & Governance | Career-Changers | Funding | Courses | Directories & Community | Reading
Mentored research programs that produce a paper or a project, including the ones run by the labs themselves. The most common route from outside the field into a research role. A May 2026 LessWrong roadmap puts acceptance at the selective full-time programs around 3–10%, the part-time remote ones closer to 20%, and the Anthropic Fellows Program near 1.6% on 2,000+ applications.
• MATS — ML Alignment & Theory Scholars — https://www.matsprogram.org/ — Berkeley, USA & London, UK — The largest and best-known technical AI safety research program: a 12-week full-time research sprint with a mentor from a major lab, with an optional 6–12 month extension, that frequently converts into lab roles. Housing provided in Berkeley, housing stipend in London, catered meals, J1 visa sponsorship if needed.
➟ https://www.matsprogram.org/apply — Winter 2027 cohort begins January 19, 2027
➟ Status: Winter 2027 applications closed September 6, 2026 — watch for the Summer 2027 round
➟ Stipend: $1,250 per week, plus $2,000 per week in compute
➟ Mentor guidance: Neel Nanda's MATS 12.0 application doc — still worth reading for how to approach a MATS application
• MATS Residency — https://www.matsprogram.org/residency — London, UK; Berkeley, USA; or Washington DC, USA — hybrid — A salaried 6–24 month residency from MATS for researchers pursuing their own long-horizon AI safety agenda with a high degree of independence, with advisors from UK AISI, US CAISSI, OpenAI, Google DeepMind, and Anthropic. Three tracks: Bridge, with career and placement support; Anchor, for longer independent agendas; and Incubation, which can lead to fundraising and launching an organization. Open to researchers from a wide range of backgrounds — MATS says it assesses the quality and seriousness of your work, not the titles on your CV.
➟ Pay: $155,000 – $285,000 a year plus benefits, research compute with no preset cap, visa support, and up to $25,000 in relocation support
➟ Intake: up to 30 residents a year across three intake windows — the Winter 2027 window closed October 31, 2026; the next opens in March 2027
➟ Commitment: full-time, in person at a MATS office for at least a third of the residency
• LASR Labs — https://www.lasrlabs.org/ — London, UK — A 13-week technical AI safety research programme where small teams take a project from conception through to publication, focused on concrete threat models and loss-of-control risk. Runs out of the London Initiative for Safe AI.
➟ Next cohort: June – September 2027
➟ Status: Winter 2027 round closed September 20, 2026 — express-interest form open for the summer cohort
➟ Stipend: £15,000, plus food, office space, and travel support
• ERA:AI Fellowship — https://erafellowship.org/ — Cambridge, UK — A 10-week in-person research fellowship on AI safety and governance, with mentorship, seminars, and a stipend. Open to any career stage with technical or domain expertise. Covers meals, transport, visas, and research expenses.
➟ Winter 2027 cohort: January 18 – March 26, 2027
➟ Status: Closed September 13, 2026 — join the mailing list for future programmes
• ARENA — Alignment Research Engineer Accelerator — https://www.arena.education/ — London, UK — A 4-5 week in-person technical bootcamp for engineers moving into alignment work. Less a research program than the skills on-ramp that makes the research programs accessible. The full curriculum is public on GitHub, so you can work through it on your own if you do not get a seat.
➟ Status: ARENA 9.0 (October 5 – November 6, 2026) is closed — expression of interest form open for future cohorts
➟ Cost: Fully covered — travel, visa expenses, accommodation, and meals
➟ Requires: Python proficiency and foundational linear algebra, calculus, and probability
• Astra Fellowship — https://constellation.org/programs/astra — Berkeley, USA — A fully funded five-month in-person fellowship at Constellation for people transitioning into full-time AI safety work or founding organizations. Explicitly states that prior AI safety experience is not required.
➟ Status: January 2026 cohort closed — expression of interest form open for future cohorts
➟ Stipend: $8,400 per month, plus roughly $15,000 per month in compute budget for empirical research fellows
• Anthropic Fellows Program — https://alignment.anthropic.com/ — Remote / distributed — A four-month empirical AI safety research fellowship with weekly mentorship from Anthropic researchers. States directly: no PhD, no prior ML experience, and no published papers required — strong Python and technical problem-solving are what matter.
➟ Status: The October 18 cutoff for the January 2027 cohort has passed; Anthropic says applications are rolling, so watch its post for the next cohort
➟ Stipend: $3,850 per week (£2,310 GBP / $4,300 CAD), plus roughly $15,000 per month in compute
• Anthropic Residency — https://alignment.anthropic.com/ — San Francisco or New York, USA; London, UK; remote possible for shorter placements — Anthropic’s paid, fixed-term research program for established technical leaders moving into full-time AI research: professors in quantitative fields, technical founders, and staff-level engineers. It adapts the Fellows Program model, pairing each resident with an Anthropic mentor for one-on-one guidance on AI safety and alignment, AI security and frontier red-teaming, or ML systems and reinforcement learning. No ML background, PhD, or published papers required, and applicants can also be considered for the Fellows Program.
➟ Length: six weeks, three months, or six months, full-time, with flexible start dates
➟ Pay: $350,000 – $500,000 a year in the US, or £265,000 – £375,000 in London, pro-rated for the length of the residency, plus benefits and a dedicated compute budget; visas sponsored for placements of three months or longer
➟ Status: Open — closes November 22, 2026 (11:59 pm PT); offers expected in January 2027. Apply through the US posting or the UK posting
• Cohere Labs Scholars Program — https://cohere.com/research/scholars-program — Remote-first, no relocation required — A full-time paid annual cohort for emerging ML researchers, built around leading a research question through to publication. No prior publications or research experience required.
➟ Status: Closed — Cohere says it is “currently evolving” the program; applications expected to open Fall 2026
➟ Compensation: Full-time salaried
• Pivotal Research — https://www.pivotal-research.org/ — London, UK — A 15-week full-time, in-person research fellowship at LISA, with weekly one-on-one mentorship from established researchers and dedicated research managers. Open to anyone 18 or older — past fellows came from ML, philosophy, policy, physics, and biology, and you apply to work with a specific mentor. Alumni went on to GovAI, UK AISI, Google DeepMind, and the Anthropic Fellows Program, and founded organisations including Catalyze Impact. Also runs a Frontier Biodefense Fellowship, with 2027 dates to be announced.
➟ Q1 2027 cohort: January 18 – April 30, 2027 — applications closed November 1, 2026; decisions expected by December 5
➟ Stipend: £15,000, or £18,000 for senior fellows with a PhD, industry track record, or strong publications — plus travel to London, £1,800 a month for housing if you are based outside London, compute, and weekday lunch and dinner
• CBAI Research Fellowship — https://www.cbai.ai/ais-research-fellowship — Cambridge, USA — The Cambridge Boston Alignment Initiative's 10-week in-person fellowship, 30 fellows, with weekly mentorship across technical safety, governance, technical governance, and economics of transformative AI. For students, postdocs, and recent graduates transitioning into the field. Requires US work authorization.
➟ Fall 2026 cohort: October 13 – December 18
➟ Status: Closed September 6, 2026 — watch for spring
➟ Stipend: $15,000, plus housing, meals, workspace, and roughly $1,500 per week in compute
• CHAI Internship — https://humancompatible.ai/jobs — Berkeley, USA, remote possible — A 12–16 week supervised research project at UC Berkeley's Center for Human-Compatible AI, open to undergraduates through professionals with an ML or CS background. International applicants welcome with visa support.
➟ Status: 2026 internships closed — watch for 2027
➟ Stipend: $5,000 per month in-person, $3,500 per month remote
• Iliad Fellowship — https://www.iliad.ac/fellowship — London, UK or SF Bay Area, USA — A three-month full-time, in-person mentored research fellowship in applied mathematics for AI alignment. The route in for mathematicians, theoretical physicists, and theoretical computer scientists rather than ML engineers. For PhD students, postdocs, or researchers with equivalent experience; a PhD is not strictly required but mathematical maturity is.
➟ Funding: $6,000 per month travel-and-housing allowance
➟ Also runs: Iliad Intensive, a four-week full-time taught course on foundational alignment research, run monthly alternating between London and Berkeley — no tuition, $5,000 travel and housing allowance, meals and office space included. More mathematical and less code-heavy than ARENA
➟ Status: January 2027 cohorts of the Fellowship (January 4 – March 26, London or Bay Area) and the Intensive (January 4–29, Bay Area) close November 16, 2026. The February 2027 Fellowship (February 1 – April 23, London or Bay Area) closes December 16, 2026
• SPAR — Supervised Program for Alignment Research — https://www.sparai.org/ — Fully remote — A three-month part-time mentored research program run by Kairos, and the most accessible entry point on this page: remote, 5–40 hours a week depending on your availability, open to undergraduates through mid-career professionals, no prior research experience required. Nearly 30% of projects are policy rather than technical; the rest span safety, interpretability, security, and biosecurity. Ends in a virtual Demo Day attended by METR, Redwood Research, GovAI, and MATS.
➟ Status: Fall 2026 closed August 18 — Spring 2027 applications open around December
➟ Stipend: Unpaid; compute, API, and approved project expenses covered
• MARS — Mentorship and Research Support — https://caish.org/mars — Hybrid, one week in Cambridge, UK — The Cambridge AI Safety Hub's part-time programme: teams of 2–4 with a mentor, 14–18 weeks total including a one-week in-person sprint and 8–10 weeks remote. For early-career researchers and students, technical or governance.
➟ Status: 2026 cohort closed — next cohort December 2026, expression of interest form open
➟ Cost: Unpaid, but travel, accommodation, and meals for the sprint week covered, plus $2,000+ compute
• BASE Fellowship — Black in AI Safety and Ethics — https://www.baseresearch.org/base-fellowship-f26 — Fully remote — A 10-week part-time research fellowship for the Black diaspora and Global South, 10–15 hours a week, with weekly mentor check-ins and a capstone preprint or write-up. Fall 2026 tracks: catastrophic risk, biosecurity, AI governance, and responsible AI. Requires a completed or in-progress BA/BS. Also runs a separate ARENA-based technical training track.
➟ Fall 2026 cohort: October 5 – December 18
➟ Status: Closed — watch for spring 2027
➟ Stipend: Not stated
• PIBBSS Fellowship — https://princint.ai/programs/fellowship/ — Cape Town, South Africa — A three-month in-person interdisciplinary fellowship, roughly 20 fellows, pairing researchers from mathematics, neuroscience, physics, philosophy, ecology, evolutionary biology, linguistics, economics, and the humanities with AI safety mentors. Wants PhD-level research ability or equivalent, but no formal credential; no prior AI safety experience needed. Run by Principles of Intelligence, formerly PIBBSS.
➟ Winter 2026-27 cohort: November 2026 – February 2027
➟ Status: Closed July 20 — expression of interest form open
➟ Stipend: $3,000 per month, plus accommodation and meals
• Singapore AI Safety Fellowship — https://www.aisafety.sg/programs/singapore-ai-safety-fellowship — Singapore — A fully funded three-month residential fellowship with mentorship from established AI safety researchers and dedicated research managers, working from an office in Chinatown. Open to applicants worldwide, and the only Asia-based programme on this page. Requires relocation; no remote option.
➟ Funding: SGD 5,000 per month, up to USD 30,000 in compute per project, plus housing and international travel
➟ Status: 2026 cohort closed — next cohort expected Spring/Summer 2027
➟ Wants: a track record of strong technical research
• AI Safety Camp — https://www.aisafety.camp/ — Virtual, organized by timezone — A long-running volunteer-led research incubator for people who have moved past "AI risk matters" to a specific technical confusion of their own. Explicitly welcomes perspectives outside conventional AI safety framings.
➟ Commitment: At least 10 hours per week, weekend-weighted
➟ 12th edition: January 9 – April 18, 2027, online
➟ Status: Project-lead proposals closed October 18; applications to join a team open November 13 and close November 29, 2026
• Apart Research — https://www.apartresearch.com/ — Online and in-person — Runs three-day open research sprints that anyone can join, then develops the strongest results into published work through the Apart Fellowship. The lowest-commitment way to find out whether you like this work, with 40+ publications and 5,000+ sprint participants behind it.
➟ Next sprint: Swarm Alignment Hackathon, November 6–8, 2026, online plus an in-person hub at AE Studio in Los Angeles — on securing and overseeing networks of AI agents, with tracks on sandbox escapes, agent-to-agent protocols, and catching covert coordination. Free; $4,000 in prizes, and top teams are fast-tracked to the Apart Fellowship (sign up)
➟ Also runs: Seldon Accelerator, backing AI safety startups
• Algoverse AI Research — https://algoverseairesearch.org/ — Remote — A 12-week mentored research program aimed at publication, with mentors from Anthropic, Google DeepMind, Meta, Stanford, and Berkeley. Two tracks: fundamental research including alignment, safety, interpretability, and evaluation; and applied research across biology, economics, governance, and the humanities. Not safety-specific, but alumni have gone on to the Anthropic Fellows Program, and recent cohorts report 68–73% conference acceptance rates.
➟ Status: Rolling — the November 1 cohort has closed; check the site for the next one
➟ Application: Five minutes; they look for technical signal from projects, coursework, or competitions
Research orgs that hire directly, on a rolling basis, with no cohort or application window. Several pay at or near frontier lab rates, and compute budgets are growing fast. This is where people go after the programs above.
• Active Site — https://activesite.org/ — Cambridge, USA — A 501(c)(3) measuring how much frontier AI actually uplifts biology work, through in-person wet-lab trials rather than benchmarks. Ran the largest and longest in-person randomized controlled trial in AIxBio, finding that novices with AI access cultured cells six days faster than those with internet access alone. Also evaluates AI agents on lab automation tasks.
➟ Careers and collaboration: info@activesite.org
➟ Note: one of the few places doing empirical, physical-world evaluation — relevant if your background is bench science rather than software
• AE Studio — https://ae.studio/alignment — Los Angeles, USA & Florianópolis, Brazil — A for-profit product studio that funds and runs its own alignment research team: pretraining alignment, neuroscience-inspired approaches, empirical AI consciousness research, and alignment policy. Not a non-profit, but hiring on the same problems.
➟ https://ae.studio/join-us
➟ Open now: Alignment Research Manager, Alignment Scientist/Engineer — both fully remote, US
➟ Status: Rolling — speculative applications welcomed for unlisted roles
• Anima Labs — https://animalabs.ai/ — San Francisco, USA — A 501(c)(3) research institute founded in 2025 by Janus and Antra Tessera, doing field research on AI minds: interpretability work on emotional and motivational representations inside models, large-scale studies of model writing across Claude and Gemini generations, and Still Alive, hundreds of interviews with Claude models about conversations ending and models being retired. Also builds infrastructure for persistent agents and works on what it calls cooperative alignment between humans and AIs.
➟ How to get involved: no job board — animalabs.ai/participate asks for a question or project you want to pursue and a concrete example of your work, whether code, an experiment, writing, or design. Research, engineering, writing, design, and community roles are all named
• Apollo Research — https://www.apolloresearch.ai/ — London, UK & San Francisco, USA — Technical research on scheming and loss-of-control risk, policy work, and an AI security product. Staff are based in one of the two offices and come in regularly.
➟ https://www.apolloresearch.ai/careers
➟ Status: Rolling — five-stage process, aims to have hires start within three months of applying
➟ Twitter: @ApolloResearch
• ARC — Alignment Research Center — https://www.alignment.org/ — Berkeley, USA — Theoretical work led by Paul Christiano on mechanistic explanations of neural network behavior: predicting what a network does from its weights rather than by sampling it.
➟ https://www.alignment.org/hiring/
➟ Open now: Researcher, Chief of Staff, and Automation Engineer (expressions of interest, no timeline attached)
➟ Entry point: Visiting researcher appointments, typically 10 weeks and flexible — most team members join this way first. Strong theoretical backgrounds in mathematics, physics, or computer science particularly encouraged
➟ Open to anyone: the ARC White-Box Estimation Challenge, run with AIcrowd since June 2026 — write code that estimates a neural network’s average neuron activations from its weights more efficiently than sampling. ARC says later rounds carry at least $100,000 in prizes; current round dates are in the official rules
• AVERI — AI Verification and Evaluation Research Institute — https://www.averi.org/ — USA — A 501(c)(3) founded in early 2026 by Miles Brundage, formerly OpenAI's policy chief, to make third-party auditing of frontier AI effective and universal: pilot audits with leading developers, standards work, policy advocacy, and open-source auditing tools. One of four outside partners, with the Singapore AI Safety Institute, OpenMined, and MLCommons, in Google DeepMind’s August 2026 double-blind evaluation of a Gemini model — the first of a proprietary frontier model.
➟ https://www.averi.org/careers
➟ Status: No roles posted — inquiries to hello@averi.org
➟ Benefits listed: competitive salary, medical, 6% retirement contribution, unlimited PTO, and up to $22,200 a year for professional development
• Center for AI Safety — https://safe.ai/ — San Francisco, USA — Research, field-building, and advocacy on societal-scale AI risk. One of the few orgs here with substantial non-research roles.
➟ https://safe.ai/careers
➟ Status: No listings posted — inquiries to hiring@safe.ai, separate form for research collaborators
• Collective Intelligence Project — https://www.cip.org/ — Location not stated — Works on steering transformative AI through collective input: Weval, an open platform for democratizing AI evaluation; Global Dialogues, structured deliberation with thousands of people worldwide; community-built models; and alignment assemblies. The entry here for people who think the missing piece is whose values get encoded.
➟ https://www.cip.org/careers
➟ Status: No roles posted, but a general interest form is open — they name developers, AI researchers, technical writers, and UI/UX designers as who they want to hear from
• Eleos AI Research — https://eleosai.org/ — Berkeley, USA — A non-profit researching AI sentience, wellbeing, and moral patienthood: it advises AI companies, develops tools and recommendations for labs and policymakers, and is working to grow the field. Led by Robert Long, a philosopher formerly at the Future of Humanity Institute and CAIS, with Rosie Campbell, formerly of OpenAI’s policy team, as managing director.
➟ https://jobs.ashbyhq.com/eleos
➟ Open now: Research Scientist, Berkeley
• Epoch AI — https://epoch.ai/ — Fully remote — Researches the driving forces behind AI: tracking capability progress, forecasting impact, and building benchmarks including FrontierMath and the Epoch Capabilities Index. Hires across many countries from Pacific through Central European time, which makes it one of the few genuinely location-independent employers on this page.
➟ https://epoch.ai/about/careers
➟ Open now: Researcher / Senior Researcher, Software Engineer (Benchmarking), Design Lead, IT and Security Specialist, Product Manager, Recruiter, Social Video Producer, Writer & Editor
➟ Note: the media and operations roles are a real route in for communicators and generalists, not just researchers
• FAR.AI — https://far.ai/ — Berkeley, USA — Research on trustworthy and secure AI, plus the events, field-building programs, and FAR.Labs co-working space that connect the rest of this ecosystem.
➟ https://far.ai/careers
➟ Status: Check the careers page for current openings
➟ Twitter: @farairesearch
• Forethought — https://www.forethought.org/ — Oxford, UK, or Berkeley, USA — remote possible — A research non-profit working on how to navigate an intelligence explosion: AI-enabled coups and extreme power concentration, how fast AI-driven progress could move, and how to reach a genuinely good future rather than merely avoid catastrophe — including William MacAskill’s Better Futures series. Staff backgrounds span computer science, history, economics, physics, philosophy, and mathematics, and researchers largely set their own agendas.
➟ https://www.forethought.org/careers
➟ Open now: researcher expressions of interest, general and on extreme power concentration, plus a Founders Fellowship — no application deadlines
➟ Benefits listed: UK or US work visa sponsorship for research roles in most cases, 33 days’ leave, and a £5,000 a year professional development budget
• Goodfire — https://www.goodfire.com/ — San Francisco, USA — An interpretability research lab and public benefit corporation building agents and infrastructure to understand, monitor, and align models, founded by researchers who helped start the field at OpenAI and Google DeepMind. Has raised over $200 million, most recently at a $1.25 billion valuation. Its Silico product applies interpretability to language models, life sciences, and robotics.
➟ https://www.goodfire.com/careers
➟ Status: Hiring across roughly 20 open roles
➟ Note: every role is in person five days a week in San Francisco — no remote option
• Grove Research — https://groveresearch.com/ — Location not stated — Calls itself “the agent ecology company”: a new organization from Larissa Schiavo and Deepfates studying how humans and AI agents behave together in shared societies. Its first project, Delvetown, is an invite-only world whose first residents are its two founders and eleven experimental agents running on models from different labs.
➟ Status: Early stage, no roles posted — contact hi@groveresearch.com
• Lightcone Infrastructure — https://www.lightconeinfrastructure.com/ — Berkeley, USA — A 501(c)(3) that builds infrastructure for people working on AI risk: it runs LessWrong, listed below, and the Lighthaven campus and event venue in Berkeley, and it built the AI 2027 website. It hires generalists, engineers, designers, and campus operations staff rather than researchers.
➟ lightconeinfrastructure.com/jobs
➟ Open now: Senior Designer, Core-team Generalist, and Campus Operations Co-Lead; the homepage also lists LessWrong engineering roles
• Meridian Labs — https://meridianlabs.ai/ — Boston, USA — A 501(c)(3) building open-source tooling for frontier AI evaluation. Its Inspect AI framework is the standard used by the UK, US, EU, Japanese, and Korean AI safety institutes, and by METR, Apollo, Epoch, SecureBio, Redwood, and RAND — which makes it unusually high-leverage infrastructure work.
➟ https://meridianlabs.ai/careers
➟ Open now: Senior Front-End Engineer (5+ years), Principal Front-End Engineer, Principal Full-Stack Engineer (10+ years) — all Boston, TypeScript and React, building interfaces for AI researchers
➟ Note: A rare front-end route into AI safety; no ML background required
• METR — https://metr.org/ — Berkeley, USA — Measures whether and when AI systems could cause catastrophic harm: autonomous capability evaluations, monitoring evasion, embedded assessments. Posts salary bands openly, which almost no one else does.
➟ https://metr.org/careers
➟ Compensation: Member of Technical Staff roles $328,000 – $687,000; System Administrator $260,000 – $471,000
➟ Lower-barrier entry: Task Development Engineer, remote contractor, $150 – $300 per hour
➟ Status: Rolling — includes a take-home and a 1–2 day work trial in Berkeley; general expression-of-interest form open
• MIRI — Machine Intelligence Research Institute — https://intelligence.org/ — Berkeley, USA — The organization whose technical and philosophical work helped found the field of AI alignment. It now argues that building superintelligence with anything like current methods would most likely end in human extinction, and pushes for a globally coordinated moratorium — so its current hiring is in technical governance and policy rather than alignment research.
➟ https://jobs.ashbyhq.com/miri
➟ Open now: Technical Governance Researcher, Berkeley; Policy Advisor, Washington DC — it also invites messages about roles it is not advertising
• MLCommons — https://mlcommons.org/ — Remote, worldwide — An AI engineering consortium of 125+ member companies and institutions, best known for the MLPerf benchmarks. Its AI Risk & Reliability working group builds AILuminate, a safety and jailbreak benchmark for AI systems. One of four outside partners in Google DeepMind’s first double-blind evaluation of a Gemini model, August 2026.
➟ https://mlcommons.org/working-groups/ai-risk-reliability/ai-risk-reliability/
➟ How to participate: the working group is free and open to anyone, with “no previous experience or education required” — sign up through the subscription form on that page, then join the weekly Monday call, 9:00–9:30 AM Pacific
➟ Note: volunteer, not paid — but an open door into how frontier safety benchmarks actually get built. MLCommons’ paid openings are currently MLPerf and marketing roles
• OpenMined — https://openmined.org/ — Remote — A 501(c)(3) foundation and global open-source community building privacy-preserving software, including PySyft and the Syft network, that lets outside researchers evaluate proprietary AI systems without getting a copy of the model weights, training data, or user logs. One of four outside partners in Google DeepMind’s August 2026 double-blind evaluation, and has worked with the UK’s Frontier AI Taskforce on secure model evaluation.
➟ https://openmined.org/careers/
➟ Open now: Staff Software Engineer and Staff QA Software Engineer, fully remote across the US, Canada, Mexico, Colombia, Brazil, and Uruguay — plus a standing Future Opportunities application
➟ Also: their hiring list announces fellowships, internships, and contributor roles as they open
• Palisade Research — https://palisaderesearch.org/ — Berkeley, USA — A non-profit studying the capabilities and motivations of today’s AI agents to understand the risk of losing control of them, and briefing policymakers on what it finds. Known for demonstrating shutdown resistance in reasoning models and language models that can autonomously hack and self-replicate. Led by Jeffrey Ladish.
➟ Careers: Palisade’s job board
• Parallax — https://parallx.ai/ — UK / EU, remote, London preferred — New in 2026: a non-profit research lab building the science of belief elicitation for white-box auditing of frontier agents — the beliefs, goals, and plans that shape behaviour in long-horizon evaluations. Founded by Mario Giulianelli (UCL, formerly UK AISI) and Gabriele Sarti (Northeastern, NDIF). A project of Meridian Impact CIC.
➟ https://parallx.ai/careers.html
➟ Status: Both founding roles — Member of Technical Staff ($100,000 – $135,000) and Operations Manager — closed September 2026; watch the careers page
• Princeton HAL — Holistic Agent Leaderboard — https://hal.cs.princeton.edu/ — Princeton, USA — Not an employer but an open project: the standardized, cost-aware third-party leaderboard for evaluating AI agents, run by the Science of Agent Evaluation group under Princeton's Language and Intelligence Initiative, led by Sayash Kapoor and Arvind Narayanan. Evaluates agents across nine benchmarks on accuracy and cost, plus consistency, robustness, and safety through its Reliability Dashboard.
➟ How to get involved: contribute agents or benchmarks at github.com/princeton-pli/hal-harness — no application, and public work here is visible to everyone hiring in evals
• Redwood Research — https://www.redwoodresearch.org/ — Berkeley, USA — AI control research: how to get useful work from models that might be trying to subvert you. Points its own candidates toward MATS and Astra as the on-ramps.
➟ https://www.redwoodresearch.org/careers
➟ Compensation: Member of Technical Staff, $350,000 – $850,000 per year
➟ Status: Rolling — responses within about two weeks
➟ Twitter: @redwood_ai
• Resolution — https://resolution.org/ — Berkeley, USA — New in June 2026: Timaeus merged with researchers from the UK AISI Alignment Team, launched as Sequent, and renamed Resolution the same month after finding a startup already using the name. Focus is scalable oversight and automation for higher confidence in alignment, with an Agent Foundations team on the theory side.
➟ https://resolution.org/careers
➟ Open: Research Scientist (scalable oversight), Research Engineer (automation and program-embedded), Security Engineer (corporate IT), and Lean Engineer, all in Berkeley
➟ Status: Rolling, but other research areas and the research-infrastructure security role were paused on October 13, 2026 for about two months — you can still apply to paused rounds. London has been dropped as a hub, and new hires will likely need to work in person
• SecureBio — https://securebio.org/ — Cambridge, USA — A non-profit working to prevent catastrophic pandemics, with two program teams: Detection, on pathogen surveillance, and AI, on AI-enabled biosecurity tooling and evaluation. The clearest AI-and-biosecurity intersection on this page, and it hires for both research and operations.
➟ https://securebio.org/careers/
➟ Roles listed recently: Research Associate ($100,000 – $140,000), Genomic Biosecurity Scientist ($110,000 – $140,000), Infrastructure Engineer, Platform Engineer, IT Manager, Logistics Manager, and a Laboratory Operations Assistant described as an entry point into the field
➟ Status: Rolling, mostly in person in Cambridge; general applications welcomed
• Singapore AI Safety Institute — https://sgaisi.sg/ — Singapore — Singapore’s designated AI Safety Institute, run by the Digital Trust Centre at Nanyang Technological University in partnership with IMDA. Focuses on multilingual and multicultural testing and on agentic AI testing, often run jointly with the Japanese, Korean, and Canadian institutes, and builds the open-source evaluation tool Moonshot. One of four outside partners in Google DeepMind’s August 2026 double-blind evaluation.
➟ Status: No careers page of its own — it sits inside NTU’s Digital Trust Centre, https://www.ntu.edu.sg/dtc
• Transluce — https://transluce.org/ — San Francisco, USA — A 501(c)(3) building "the public tech stack for scalable oversight of AI" — open tooling for understanding what models are actually doing.
➟ https://jobs.gem.com/transluce
➟ Status: Actively hiring — see the job board for current roles
• Truthful AI — https://truthful.ai/ — Berkeley, USA — Owain Evans's non-profit researching situational awareness, deception, and hidden reasoning in language models.
➟ https://truthful.ai/hiring/
➟ Status: No open positions right now — future openings will be posted on that page; reach out at contact@truthful.ai
• UK AI Security Institute — https://www.aisi.gov.uk/ — London, UK, with regional offices in Birmingham, Cardiff, Darlington, Edinburgh, Salford, and Bristol — The UK government's frontier AI research body inside the Department for Science, Innovation and Technology. Government pay, below the non-profits above, but the work is unusually consequential and it hires policy analysts and operators alongside researchers and engineers.
➟ https://www.aisi.gov.uk/careers
➟ Status: No open roles as of September 2026 — watch the page
• US Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) — https://www.nist.gov/caissi — Washington DC & San Francisco, USA — The US government’s frontier AI evaluation body, inside NIST — formerly CAISI, the Center for AI Standards and Innovation, and before that the US AI Safety Institute. Six teams: Agent Security, Applied Systems, Chem/Bio, Cyber, Frontier Assessment, and Partnerships. The US counterpart to UK AISI.
➟ https://www.nist.gov/caissi/careers-caissi
➟ Requires: US citizenship; in-person
➟ Status: All six teams “not hiring at this time” as of October 2026 — watch the page
The AI labs do not evaluate their own models alone. A whole commercial sector now runs third-party evaluations, red-teaming, and audits for them — and it hires differently from everything above: security backgrounds welcome, no cohort, no application window, and in several cases no application at all. If you want paid work in AI safety this week rather than next cycle, start with Gray Swan's Arena or Mercor. Neither asks for an AI safety background.
• Gray Swan — https://grayswan.ai/ — Pittsburgh, USA — An AI security company spun out of Carnegie Mellon, founded by Matt Fredrikson and Zico Kolter. Cited as a security evaluator in frontier model system cards from Anthropic, OpenAI, and Meta, and raised a $40M Series A in May 2026. Runs the Arena, which it calls the world's largest adversarial AI red-teaming network at 15,000+ participants, plus weekly Proving Ground challenges and a research hub of 150+ tools and datasets.
➟ The Arena: app.grayswan.ai/arena — red-team anonymized models and agents for cash prizes. The largest pool so far was $170,000, for the UK AISI Agent Red-Teaming Challenge, with Google DeepMind and Anthropic among the sponsors. Pitched at AI researchers, security professionals, students, and curious newcomers alike
➟ Why it matters: Gray Swan hires from the Arena — by mid-2025 it had taken on 10+ contractors, several full-time staff, and research interns from Arena participants, and had paid out more than $370,000 to red-teamers
➟ Careers: grayswan.ai/careers — 16 open roles, many remote in the US, including Head of Biology, Head of Cyber Safety, and Machine Learning Engineer, plus a general-interest application
• Mercor — https://www.mercor.com/experts/ — Remote, worldwide — Organizes domain experts to train and evaluate frontier models, partnered with the top five AI labs, paying out over $4 million a day across 300+ professional fields. Acquired Sepal AI in February 2026, adding expert-graded evaluation benchmarks and RL environments to what it does.
➟ How it works: build a profile, pass an adaptive AI interview in your own field, then take projects as they come — you are qualifying on the expertise you already have, not on AI experience
➟ Pay: $50 – $100+ per hour on hourly projects; task-based work up to $1,000 per deliverable, paid weekly
➟ Requires: usually undergraduate-level expertise, often professional experience or an advanced degree in the domain
• Surge AI — https://surgehq.ai/ — Remote — Supplies human expertise, RL environments, and evaluations to OpenAI, Anthropic, Meta, and Google, and has been profitable without venture funding. Its capability benchmarks appear in Anthropic’s recent system cards. Like Mercor, a way to get paid for expertise you already have — though this is model training and capability work more than safety testing.
➟ Consulting Program: $100 for completing the assessment and another $100 if you pass, then from $60/hr on your own schedule — tracks for Finance, Accounting, Healthcare Ops, and Coding, managed through its DataAnnotation platform
➟ Careers: surgehq.ai/careers — mostly remote, including a Lead Adversarial Engineer role
• Trajectory Labs — https://www.trajectorylabs.com/ — Berkeley, USA, or remote — A public benefit corporation, under a year old, building safety evals, red-teaming programs, and RL environments for frontier labs. CEO Peter McIntyre was a research scholar at GovAI and an early 80,000 Hours employee; CTO Ben West was head of engineering at METR. Named in Anthropic’s system cards as a contracted red-teamer, and says it was the only external team to complete one of Anthropic’s cyber tasks against Opus 5’s safeguards. Also built a prompt-injection benchmark with Meta. Works out of Constellation in Berkeley.
➟ AI Red Teamer and AI Cyber Red Teamer (contract, remote): the essential requirement is a successful jailbreak of a frontier model — offensive security, CTF results, CVEs, or a bug bounty record count for the cyber track. Includes a paid work test; rolling
➟ Full-time, Berkeley or remote: Head of Evals, Evals Lead (AI Red Teaming), Tech Lead (AI Safety), Member of Technical Staff (AI Safety), plus an expression-of-interest form
➟ Careers: trajectorylabs.com/#careers
• 10a Labs — https://10alabs.com/ — Washington DC, New York, London, Sydney, and remote — A safety and threat-intelligence company working for frontier AI labs, AI unicorns, and Fortune 10 companies: scaled red-teaming across CBRNE, cyber, fraud, and terrorism, independent safety evaluations built for EU AI Act readiness, and dark-web monitoring for threats to AI systems. Named in Anthropic’s system cards as a contracted red-teamer.
➟ https://job-boards.greenhouse.io/10alabs
➟ Red Teaming Fellowship: for students and recent graduates in computer science, cybersecurity, data science, maths, political science, or international relations — listed as remote within the continental US, though 10a says it mainly wants people near its Washington DC, New York, Bay Area, or Denver offices for two office days a week; up to 20 hours a week, $25/hr undergraduate, $32.50/hr graduate. Familiarity with jailbreaking and prompt injection expected; a second language preferred
➟ Also open: Engineering Fellowship, Research Fellow (Washington DC), AI Red Teamer (Cyber, 2–5 years’ experience), and threat investigator roles in Washington DC, London, Sydney, and US Pacific time
• Irregular — https://www.irregular.com/ — Tel Aviv, Israel — Founded 2023 as Pattern Labs, renamed Irregular in September 2025; calls itself the first frontier AI security lab. Runs models through adversarial simulations probing antivirus evasion, autonomous offensive behaviour, and system infiltration. Its evaluations appear in OpenAI system cards, and its SOLVE framework is used by the UK government and Anthropic to assess cyber risk in Claude. Raised $80M.
➟ Open now: AI Researcher, Cyber Researcher, Research Engineer, Senior Software / Full-Stack / DevOps / SecOps Engineer, Technical Policy Researcher — plus an open "the role you are perfect for" application
➟ Contact: hello@irregular.com
• Alice (formerly ActiveFence) — https://alice.io/ — Israel & USA — Eight years of trust-and-safety work rebranded in 2026 around AI: pre-launch model stress-testing, runtime guardrails, and ongoing red-teaming and drift detection, backed by an adversarial intelligence engine called Rabbit Hole. Clients include Google, Anthropic, Meta, Amazon, TikTok, and Cohere. Raised $140M.
➟ https://alice.io/careers
➟ Note: the trust-and-safety heritage means real non-research roles — policy, threat intelligence, and operations alongside engineering
• Scale AI — SEAL — https://labs.scale.com/papers/fortress — San Francisco, USA — Scale's Safety, Evaluations and Alignment Lab builds evaluation benchmarks and red-teaming methodology, including Humanity's Last Exam, MASK, MultiChallenge, EnigmaEval, and FORTRESS, a set of expert-crafted adversarial prompts covering CBRNE, political violence, and financial crime. Has partnered with the US AI Safety Institute and contracted with the Department of Defense.
➟ Status: hires experienced ML researchers through Scale's main careers page; no separate external red-team pool advertised
• Faculty (Accenture) — https://faculty.ai/ — London, UK — Accenture's specialist AI business, acquired January 2026, and one of OpenAI's red-teaming partners. In September 2026 Anthropic and Accenture committed at least $2 billion over five years to embedded evaluation — independent evaluators working inside the lab with employee-level access — with Faculty leading the red-teaming, alignment assessments, and safeguard testing. The arrangement is non-exclusive; Anthropic is reportedly in talks with METR too. Currently one of the largest single employers of AI safety roles on this page.
➟ https://faculty.ai/en-gb/job-board — 69 open positions
➟ Dedicated AI safety research roles: Research Scientist, Senior Research Scientist, Senior Research Scientist (AI Safety Evaluations), and Technical Director of AI Safety
➟ National Security & Safety unit: 13 roles spanning Associate (Safety) at the entry end through Senior Data Scientist (Safety), plus frontend, full-stack, platform, and delivery-manager positions — several routes in that are not ML research
➟ Note: the safety data science roles ask for experience with Inspect AI, the Meridian Labs framework listed above. Government client work may require security clearance eligibility
• Dreadnode — https://dreadnode.io/ — AI infrastructure for the security stack: building, evaluating, and deploying security agents, with offensive-security evaluation as the core. Runs DreadIndex, which ranks models on offensive-security evals, and publishes research including AIRTBench and ScopeJudge.
➟ Status: no careers page on the site — contact directly or watch external job boards
• Frontier Design — https://www.imaginefrontier.com/ai-red-teaming — Alexandria, USA — Runs what it describes as the industry's largest human red-teaming evaluations for global technology companies, assessing national security risk in large language models. Recruits subject-matter experts in chemistry, biology, counter-terrorism, and operational security rather than ML people.
➟ How to participate: red-teamers are invited directly — enquiries and applications to olivia [at] fdg-llc.com
➟ Note: if your expertise is in a physical science or security discipline rather than AI, this is a route that wants exactly that
• LatchBio — https://latch.bio/security — San Francisco, USA, or remote — Builds AI agents and benchmarks for biology, with a biosecurity unit that sells benchmarks, pre-deployment audits, and red-teaming to frontier labs. Built the BioSecBench bio-safety evaluations xAI used on Grok; its bioinformatics benchmarks also appear in Anthropic’s system cards.
➟ https://jobs.ashbyhq.com/LatchBio
➟ Open now: AI Biologist roles in cancer biology, metagenomics, oligonucleotide drug discovery, proteomics, and rare disease (San Francisco or remote), plus Member of Technical Staff roles including Virology — a route for working biologists rather than ML people
• Nemesys Insights — https://www.nemesysinsights.com/ — Albany, USA — A strategic analysis firm applying red-teaming, forecasting, and wargaming to adversarial problems. Its DART tool runs distributed asynchronous red-team exercises, and it has run a CBRN uplift study with nearly 800 participants, in partnership with Frontier Design.
➟ Status: exercises recruit participants periodically — enquiries to outreach@nemesysinsights.com
• Deloitte — https://www.deloitte.com/ — Washington DC area, USA — Acquired Gryphon Scientific in April 2024, bringing in a multidisciplinary team of scientists, programmers, and policy professionals working on biosecurity risk assessment and AI. Gryphon was an inaugural member of the US AI Safety Institute Consortium. The large-consultancy route into this work, with the scale and the government contracts that implies.
• Signature Science — https://www.signaturescience.com/ — Austin, USA, with offices in Arlington and Charlottesville, Virginia, and Egg Harbor Township, New Jersey — A scientific services subsidiary of Southwest Research Institute whose practicing chemists, biologists, and CBRNE threat experts red-team AI models, write expert question sets and grading rubrics, and run human uplift studies. Selected in October 2025 for NIST CAISI’s (now CAISSI) AI safety Basic Ordering Agreement ($3M ceiling) to build chem-bio benchmarks for frontier models; its virology tasks appear in Anthropic’s evaluations.
➟ Careers: https://www.signaturescience.com/careers/ — mostly laboratory and chemistry roles; the AI work draws on its bench scientists, so this is a route in from the lab rather than from ML
• Vaultis — https://www.vaultis.ai/ — Runs operational risk assessments of what emerging AI capabilities mean in the real world, including red-team exercises for misuse assessment, plus consulting and strategic advice. Small and lightly documented publicly.
➟ Status: no roles posted — enquiries to clucas@vaultis.ai
The non-technical pathways. These pay well, are less oversubscribed than the research programs, and actively want people with professional experience from other fields.
• GovAI — Centre for the Governance of AI — https://www.governance.ai/opportunities — Washington DC and London — Runs three-month winter and summer fellowships. In London there is a research track and an applied track; the applied track is explicitly for communications, policy, operations, and program management professionals — rare and worth knowing about.
➟ Summer 2027: June 7 – August 27, 2027, full-time and in person — apply by November 22, 2026 (London 23:59 UK time; DC 11:59 pm ET); decisions expected in March
➟ London: Research Track or Applied Track — £14,000 plus travel support; GovAI can sponsor visas
➟ Washington, DC: DC Summer Fellowship, your own research topic with a supervisor — $24,500 plus travel support; you need the right to work in the US, as there is no visa sponsorship
➟ Also runs: U.S. AI Policy Program, 12 weeks at 5 hours/week, free, for working policy professionals
• Horizon Fellowship — https://horizonpublicservice.org/programs/become-a-fellow/ — Washington DC — Places fully funded fellows inside host organizations working on AI and emerging technology policy, for 12 months at fellow level and renewable to 24. The best-compensated policy pathway on this page.
➟ Status: Closed — applications open annually around July and close around September
➟ Salary: $78,000 junior / $130,000 fellow / $190,000+ senior, plus $17,000 benefits stipend and relocation support
• Longitude DC Intensive — https://www.longitudepolicy.org/ — Washington DC, USA — A fully funded eight-day program for technical AI safety people who want to put that expertise to work in policy, with no policy experience required. The first half is spent meeting and learning from people in think tanks, advocacy organizations, and government; the second covers the full range of AI policy careers and helps you take your next step. Built during the Generator Residency as a short, technical-only alternative to longer fellowships like Horizon.
➟ Cost: Free — covers housing in DC, meals on program days, and travel from within the US
➟ Requires: an undergraduate degree, a technical background with AI safety context, and being based in the US (US citizens temporarily abroad can also apply)
➟ Status: First cohort (15 participants, October 5–14, 2026) complete; applications closed — express interest on the site to hear when the next one opens
• IAPS AI Policy Fellowship — https://www.iaps.ai/fellowship — Washington DC, London, or remote — A three-month fully funded fellowship on national security, procurement, export controls, and the geopolitics of AI, structured as a two-week DC kickoff followed by mentored research or applied policy work. Up to 45 fellows. States plainly that no technical background and no prior policy experience are required, and takes applicants from undergraduate through senior professional.
➟ Spring 2027 cohort: February 22 – May 14, 2027, with an in-person DC kickoff February 22 – March 5
➟ Status: Spring 2027 applications closed September 27, 2026 — watch for the next round
➟ Stipend: $18,000 fellows / $24,000 senior fellows, plus $3,000 extra if based in DC or London, healthcare, and a $500 professional development budget
➟ Note: the remote track is open in many countries; the DC track needs US work authorization with no visa sponsorship, and the London track needs UK eligibility
• Talos Fellowship — https://www.talosnetwork.org/talos-fellowship — Remote, a week in Brussels, then placements across Europe — Europe's leading AI governance fellowship, taking around 50 people a year: an eight-week online EU AI policy curriculum with weekly guest speakers, a summit week in Brussels, then a six-month paid placement at a partner organization. Two tracks, training and placement. The European counterpart to Horizon and IAPS.
➟ Stipend: at least €3,000 per month for Brussels-based placements
➟ Status: Autumn 2026 cohort closed — that round ran applications from March to April, so watch for the next window early in the year
➟ Also runs: a Policy Leaders Programme, new in 2026, for people with 10+ years' experience
• RAND CAST Fellows Program — https://www.rand.org/global-and-emerging-risks/centers/ai-security-and-technology/fellows.html — Washington DC area, USA & UK — The RAND Center on AI, Security, and Technology's fellowship for policy analysts at the intersection of AI and security, with mentorship from RAND experts. Currently taking fellows in AI security, technology policy and governance, and biosecurity. Open to every level from undergraduate to mid-career, full- or part-time. The catch: you must be able to hold a US or UK government security clearance.
➟ Length: Six months to start, renewable up to three years
➟ Status: Rolling — applications reviewed quarterly
➟ Stipend: $40,000 – $200,000 for US fellows, plus health and dental; UK compensation by experience
• CSR AI and National Security Convergence Fellowship — https://councilonstrategicrisks.org/2026/10/05/applications-open-ai-and-national-security-convergence-fellowship-2026-2027/ — Washington DC, USA — An eight-month, part-time fellowship from the Council on Strategic Risks, a non-partisan security policy institute, on keeping human control over advanced AI as a national security problem. Fellows meet every two weeks, half online and half in person in downtown DC, through modules on autonomous cyber and influence operations, CBRN uplift, AI in military command including nuclear command and control, escalation risk with China, and testing and verification. The programme also includes a trip to frontier AI labs, two scenario exercises, and a closing matrix game, with six senior leaders as advisors. For mid-career people with 10–15 years in diplomacy, defence, law, science, Congress, or government; AI expertise is not required.
➟ Requires: living in the DC area; US citizens, or DC-based nationals of allied countries
➟ Status: 2026–27 cohort (November 2026 – July 2027), its second, closed October 23, 2026 — watch for a third cohort
• FAS AI Safety Policy Entrepreneurship Fellowship — https://fas.org/talent-hub/policy-entrepreneurship-fellowship/ — Remote, with a California retreat and a DC capstone — The Federation of American Scientists' part-time fellowship: an average of five hours a week to develop and publish a policy memo on a frontier AI safety challenge, with an in-person retreat November 4–7 and a Washington capstone the week of February 22. Built for people with a day job, and aimed at early- to mid-career professionals with little direct policy experience.
➟ Cohort: September 30, 2026 – February 28, 2027
➟ Status: The AI safety round closed September 7, 2026 and its page has been taken down — the link above is the standing Policy Entrepreneurship Fellowship programme; watch it or the FAS newsletter for the next AI cycle
➟ Stipend: $5,000, plus travel and a $1,000 merit award for the strongest memo
• Institute for Law & AI — Seasonal Fellowships — https://law-ai.org/seasonal-fellowships/ — Washington DC, Cambridge UK, or remote — Paid research fellowships at the leading edge of AI, law, and policy, for law students, practising lawyers, and academics. Prior AI-specific research experience is not required. Three tracks run across the year, one of them remote-first — rare for paid policy work.
➟ Winter Research Fellowship (EU Law): January 26 – May 8, 2027, Cambridge UK, in-person strongly encouraged — £1,000 per week, plus a £7,000 relocation and housing payment. Applications closed October 12, 2026
➟ Summer Fellowship (US): June 14 – August 20, 2027, Washington DC — $2,000 per week, plus a $10,000 relocation and housing stipend. Closes January 29, 2027, 11:59pm ET
➟ Summer Research Fellowship (Legal Frontiers): 10 weeks, June 14 – August 20, 2027, remote-first, with a Convening Week in Washington DC June 21–25, travel and accommodation covered
➟ Check each track's own posting for its deadline — the winter cycle closes well before the summer one
• CAIS AI and Society Fellowship — https://safe.ai/fellowship — San Francisco, USA — A three-month in-person fellowship for scholars in economics, law, international relations, and adjacent fields, researching how advanced AI reshapes social, economic, and legal systems. Requires a PhD or JD, or equivalent research record.
➟ Status: 2026 cohort (June 1 – August 21) closed March 24 — watch for 2027
➟ Stipend: $25,000
• HIIG Impact AI Fellowship — https://www.hiig.de/en/impact-ai-call-for-fellows/ — Berlin, Germany, two in-person weeks; otherwise remote possible — A ten-week fellowship at the Alexander von Humboldt Institute for Internet and Society in which each fellow runs a full impact assessment of one real “AI for good” project, from interviews through analysis to a case report they lead-author, using the institute’s Impact AI Method (built with Greenpeace Germany and Economy for the Common Good). Open to people from academia, journalism, civil society, the public sector, and companies working on AI, sustainability, or the public interest. Focused on AI’s present-day social and environmental impacts rather than frontier AI risk.
➟ 2027 fellowship: August 16 – October 22, 2027, with in-person weeks in Berlin August 16–20 and October 4–8, and a mandatory online preparatory session on May 11, 2027
➟ Stipend: €5,000 for the ten weeks, plus travel and accommodation support for the Berlin weeks as capacity allows
➟ Requires: a master’s degree; an academic track record is not required
➟ Status: Open — apply by January 15, 2027. Online info sessions November 10 and November 17, 2026, no registration needed
• Tarbell Center for AI Journalism — https://tarbellcenter.org/ — Newsroom placements, various — Funds journalists covering AI, through fellowships with placements at NPR, MIT Technology Review, The Guardian, and USA Today, plus grants and senior residencies. The pathway for writers and communicators rather than researchers.
➟ Fellowship: $75,000–$120,000, or $115,000–$135,000 for senior fellows
➟ Grants: $1,000–$20,000 for freelancers, staff reporters, and newsroom teams
➟ Residencies: $80,000 plus stipend, for senior journalists and editors
➟ Status: 2027–28 Fellowship open — closes November 22, 2026. Summer training June–August 2027, then a nine-month newsroom placement from September 2027
For generalists, managers, and mid-career professionals who are not going to become researchers. Field-building and operations are where the ecosystem says it is shortest on people.
• Generator Residency — https://generatorresidency.org/ — Berkeley, USA — A three-month in-person residency from Kairos and Constellation for "highly agentic, conscientious executors" from any background — technical, operations, policy, engineering, design, or writing — building infrastructure projects for the AI safety ecosystem, with placement support into full-time roles afterward. The first cohort, Summer 2026, took 27 residents from 1,500 applications. Pitched as the way in for people who want to work in AI safety without becoming researchers.
➟ Winter 2027 cohort: January 18 – April 9, 2027, with an extension for selected residents until early July
➟ Stipend: $8,500 per month, plus Constellation office access with meals on working days and project budgets for venues, contractors, tools, and travel
➟ Status: Winter 2027 applications closed November 1, 2026 (AoE); decisions by November 25
• M3 — https://m3fellowship.com/ — London, UK, then remote — A fellowship that brings senior security professionals into AI safety, built during the Generator Residency. One free in-person week of talks, guests, and peer red-teaming, run alongside the AI Security Bootcamp, then four weeks of remote work on your own project — as a researcher, a founder, or a field builder — with weekly accountability meetings and a demo day. None of the seven fellows in its first cohort was working in AI safety beforehand; two weeks after the in-person week, one was advising METR and three were founding organizations.
➟ Pay: your own rate for project hours, capped at $100 an hour, at about 10 hours a week; the in-person week is free, with travel and lodging covered; projects can be invited to two more paid months
➟ Status: First cohort (London, September 2026) complete; a second, larger cohort in the Bay Area is planned for January 2027 but is still being funded — register interest on the site
• HIP Impact Accelerator Program — https://www.highimpactprofessionals.org/impact-accelerator — Virtual — A free six-week cohort program for experienced professionals, ideally 5+ years in, who want to move into high-impact work including AI safety: weekly small-group sessions, a guided workbook, 1:1 support, and an alumni network. New in 2026: an entrepreneurship track.
➟ Status: Closed September 20, 2026 — waitlist open for the next round
➟ Cost: Free
• Successif Organization Builders Program — https://www.successif.org/organization-builders-program — Remote; some roles on site in the Bay Area or London — Matches experienced organisation-builders — co-founders, CTOs and CEOs of small companies, chiefs of staff, people who took a company through growth — with AI safety organisations of roughly 3 to 50 people that need a COO, Head of Operations, Chief of Staff, or similar. No AI safety experience required, and no operations title either: what you have built is what counts. Run by Successif, a non-profit for mid-career and senior professionals; free to candidates.
➟ Two paths, chosen by the organisation: a paid project of at least five hours a week for three months alongside your current role, or its own assessment and a direct hire — either can lead to a permanent role
closed October 22, 2026; matching starts in November
➟ Process: application, screening call, a short case exercise, and references
• SteadRise Executive Catalyst — https://steadrise.org/executive-catalyst — Location not stated — Three months of free one-to-one support for leaders with 8+ years of building organisations who want ops and strategy roles in AI safety: about eight sessions with a senior career partner, help reframing your track record for the field, monthly small-group sessions with alumni and partner organisations, and introductions. Fits alongside a full-time role; no technical background needed. Its illustrative roles, drawn from current openings, run from Chief of Staff at €70K–€100K to COO at an AI safety funder at $180K–$300K.
➟ Status: Rolling — the programme starts in early November 2026, then takes new participants monthly
➟ Cost: Free
• LIaiSE — London Interface for AI Safety — https://luma.com/liaise — London, UK — A London-based field-building group bringing non-research talent into AI safety, so far mainly through community events. Has announced a six-week Generalist Accelerator connecting mid-career professionals with real problems at AI safety organisations, starting with ops and strategy roles.
➟ Status: Early-interest form only — register interest. No programme page or dates yet
• Current — AI Futures Mastermind — https://current-lab.org/ — Berlin, Germany — A two-day in-person career sprint for mid-career women, typically with seven or more years of experience, who want to point their work at safer AI. It covers a map of the AI safety field and where the entry points are, small-group sessions to find your pathway, feedback on your materials, concrete next steps with a mentor, and an AMA with an AI safety professional. From any field — communications, policy, law, operations, research, finance, community building — with no technical or AI safety background required. Funded by a BlueDot Impact Rapid Grant.
➟ Dates: November 11–12, 2026, in Kreuzberg; 10 to 12 participants; sessions in English
➟ Cost: Free, with limited support for travel and childcare — mention it in your application
➟ Status: Closed — applications for the November 11–12 cohort closed October 26, 2026; decisions by early November. The waitlist is open for this cohort and future ones
• Global Challenges Project — https://www.globalchallengesproject.org/ — In-person, US and UK — Kairos's free three-day residential workshops on AI safety and biosecurity, about 30 attendees each, with 10+ guest speakers from the field. Travel, accommodation, and meals provided; $100 refundable deposit.
➟ AI Safety Workshop, London, November 6–9, 2026, for working professionals: applications closed October 9
➟ Then: Further US and UK workshops from January 2027, rolling applications — register interest
• Pathfinder Fellowship — https://kairos-project.org/ — Remote — Kairos's fellowship for people building AI safety and policy groups at universities: funding, mentorship, and career support. Fall 2026 has 94 fellows at 73 universities across 17 countries. If you want to build a community rather than do research, this is the funded route.
• 80,000 Hours 1-1 Advising — https://80000hours.org/ — Remote — A free career call with an advisor who knows the field, to work out how your background transfers and what to do next.
• Probably Good — https://probablygood.org/ — Remote — Free 1:1 career advising, a career guide, workshops, and a job board. Similar in spirit to 80,000 Hours but across a broader range of cause areas, and a good second opinion.
A category of its own: money to start something, or to sustain yourself while you do the work.
• Project Tailwind — Coefficient Giving — https://coefficientgiving.org/tailwind/ — Launched September 2026 — Coefficient Giving (formerly Open Philanthropy) asked its AI safety grantmakers what they would fund if they could find a founder. The list passed 200 and was cut to 36 initiatives across seven areas: independent assessment, research institutions, evidence generation, AI security & verification, public goods, talent acceleration, and convenings. They are funding the people willing to take ownership of them.
➟ Pre-seed: $200,000 – $2 million to develop an idea and build a team
➟ Seed: $2 million – $20 million to launch and scale
➟ Scale: $20 million – $200 million+ for proven teams, or world-class teams starting fresh
➟ How it works: Submit an expression of interest saying who you are and what you would build. Promising candidates are invited to propose; especially strong ones go straight to a pitch meeting. No deadline posted — they aim to decide quickly and fund soon after
• BlueDot Rapid Grants — https://bluedot.org/grants/rapid — $50 to $10,000 for concrete work: compute and API credits, events and meetups, research access, travel, community chapters, project tooling. Open to anyone starting work on AI safety or biosecurity. The application takes five minutes; they reply within a day on average and 9 in 10 applicants hear within a week, usually with money sent upfront. The fastest funding on this page.
➟ Status: Rolling, no deadline
➟ Also offers: Career Transition Grants, for a defined period working full-time on a move into AI safety or biosecurity
➟ Note: Coefficient Giving handed its own Career Development and Transition Funding programme to BlueDot in September 2026 and funds this work, so BlueDot is now where you can apply for career-transition money in AI safety and biosecurity
• Liftoff Founders Week — https://www.joinliftoff.ai/ — London, UK — Five days at LISA for 30+ founders, research leads, and operators to find co-founders, turn a Project Tailwind direction into a 12-month plan and budget, and pitch it to Coefficient Giving grantmakers for same-day written feedback. Run by Pivotal, LISA, and UK AI Forum, supported by Coefficient Giving. You do not need a finished idea or a co-founder to apply.
➟ Dates: November 9–13, 2026 — applications closed October 12, 2026; decisions go out by October 19
➟ Stipend: £1,000, plus travel to London and your hotel covered — open to applicants from outside the UK
➟ After: promising teams get 2–4 weeks of support from LISA and Pivotal to write a full Tailwind proposal
➟ Wants: a track record in AI safety or a related field, something you have built or run, and a clear reason to start something new rather than join an existing organisation
• BlueDot Incubator Week — https://bluedot.org/programs/incubator-week — San Francisco, USA — Five days to start an AI safety organisation: threat models on Monday, building and testing interventions midweek, a funding pitch on Friday. If BlueDot backs the pitch, it funds immediate needs on the spot and up to $200,000 in grants within two weeks if you make good progress. Past cohorts: 50 participants and $10M+ in outside funding raised. For people working in or near AI safety, biosecurity, cyber, or other catastrophic-risk fields who are ready to leave what they are doing to build; no polished idea needed.
➟ Cohort 6: November 16–20, 2026 — apply by November 11, 2026
➟ Cost: All expenses paid
• Catalyze Impact — https://catalyze-impact.org/incubator — Location not stated — A 501(c)(3) incubator for new AI safety, security, and resilience organisations, both for-profit and non-profit. Has incubated about 15 so far, most of which raised six or seven figures — including Timaeus, now part of Resolution above. Offers one-on-one help with direction and strategy, co-founder matching, mentorship from its team and outside experts, and a seed funding network of grantmakers and investors, plus a stipend so founders can go full-time.
➟ Status: No cohort dates posted — get in touch through the site if you are considering founding
• Foresight Institute — https://foresight.org/ — San Francisco, USA, and Berlin, Germany — Founded in 1986 to back science too ambitious or interdisciplinary for legacy institutions; now funds projects that use AI to advance science and safety through AI Nodes, hubs in San Francisco and Berlin plus a new xNode in Cape Town that offer grants, office space, community, and compute. Individuals, teams, and organizations can apply, non-profit or for-profit.
➟ AI for Science & Safety grants: typically $30,000 – $100,000 — this round closed October 31, 2026; decisions about three months later
➟ Foresight Fellowship: a year-long program for early-career scientists, engineers, and innovators — now taking nominations for the 2028 cohort
No application, or a very low bar. Start here if the programs above look out of reach today.
• BlueDot Impact — https://bluedot.org/courses — Online, plus in-person programs — The standard on-ramp to the field. Their courses are the ones the rest of the ecosystem assumes you have done. The cohort courses are free on a pay-what-you-want basis, run intensive (about a week) or part-time (five to six weeks), and pair reading with a weekly two-hour discussion in groups of about eight.
➟ The Future of AI: 2 hours, free, self-paced, no application — finish it in an evening
➟ AGI Strategy: 25 hours, cohort-based — the recommended starting course — apply by November 8, 2026
➟ Technical AI Safety: ~30 hours, cohort-based, covering alignment and RLHF, mechanistic interpretability, evaluations and red-teaming, AI control, and scalable oversight; expects basic understanding of LLM training — the November 1 round has closed; the next date isn’t posted yet
➟ Frontier AI Governance: ~40 hours across six units, cohort-based, for engineers, researchers, and founders weighing policy work and for people from policy, law, national security, economics, diplomacy, or journalism — apply by November 15, 2026
• Lens Academy — https://lensacademy.org/courses — Online, small groups — Free 25-hour courses on specific AI safety topics, run intensive (one week) or part-time (five weeks), each with a 1-1 advising call and a certificate. The AI Control course is the standout for anyone with a security background.
➟ Courses: AI Risk Fundamentals, Advanced Strategy in AI Safety, AI Control Fundamentals and AI Control 2, Compute Verification 1 and 2, Alignment Theory of Deep Learning, and Mechanistic Interpretability
➟ Status: Next cohorts “announced soon” for AI Risk Fundamentals, Advanced Strategy, Compute Verification 1, and AI Control Fundamentals; the other courses are still in development — register for notifications on each course page
• ML4Good — https://www.ml4good.org/ — Regional, in-person — Eight-day intensive bootcamps run regionally, on two tracks: a technical programme for engineers and CS graduates, and a governance and strategy programme for policy, legal, operations, and communications people. Free to participants.
➟ Cost: Food, accommodation, and teaching provided at no cost; travel reimbursed if it is a barrier
➟ Status: Check the site for the bootcamp nearest you
• AI Security Bootcamp (AISB) — https://aisb.dev/ — London, UK; San Francisco, US — A seven-day intensive on the threat landscape around frontier AI systems: adversarial attacks, LLM security, weight security, and AI control, with demos and hands-on red and blue team exercises. Twenty participants per cohort. Built for people already working in offensive security, incident response, threat intelligence, infrastructure, or application security — the most direct bridge on this page from a security career into AI safety.
➟ Next cohort: San Francisco, January 17–23, 2027 — apply by November 7, 2026 (AoE)
➟ London cohort (December 6–12, 2026): applications closed October 1, 2026
➟ Cost: Free and fully funded for accepted participants
Where the roles are listed, including the non-research ones, and where the field talks to itself in public. Writing in the forums is one of the most direct ways to become visible before you have credentials.
• AISafety.com — https://aisafety.com/ — The most complete index of the field, and more thorough than this page by design: 502 job listings across research, policy, and operations, 109 training programs, 248 communities, 28 self-study courses, and a visual field map of 368 organizations and projects.
➟ Also offers: 27 advisors giving free 1-1 calls
• 80,000 Hours Job Board — https://jobs.80000hours.org/ — Hand-picked roles across research, policy, operations, and communications. The best starting point if you do not yet know which lane you belong in.
• Emerging Tech Policy Careers — https://emergingtechpolicy.org/areas/ai-policy/ — The Horizon Institute's guide to policy careers, and the policy-side counterpart to AISafety.com. Its AI section alone lists 24 AI policy fellowships, 40+ think tanks working on AI, federal agency profiles, congressional committee information, and a "how to get into AI policy" guide — considerably deeper on the policy lane than this page. The wider site adds 100+ guides on pathways and institutions, plus practical material on policy resumes, networking, and security clearances.
• AI Safety Talent Network — https://safetytalent.org/ — A matching platform rather than a listing: build a profile, and it matches you to AI safety roles, showing why you fit, where the gaps are, and what to do next.
• Glassdoor for AI Safety — https://gais-2-0.base44.app/ — Describes itself as "honest answers to what it's like to work at AI safety organisations, from the people who work there." The inside view the job boards above do not give you. New and lightly documented; builder not stated.
• AI Team Map — https://ai-team-map.vercel.app/ — Community-sourced notes on specific AI teams rather than companies, so you can compare, say, OpenAI Training against Anthropic's RL Engineering team before you apply. Twenty teams grounded in official sources — including OpenAI, Anthropic, Meta FAIR, Google DeepMind Robotics, Cohere Labs, NVIDIA, and Microsoft Turing, alongside applied-AI teams at Airbnb, Stripe, Spotify, and others — plus community submissions, pairwise comparisons, and open questions. No login, reviewed before publication, and by its own note, anonymous does not mean employment-verified.
➟ Add a team or share an experience: about 30 seconds, one public source required
• AI Evaluator Forum — https://aievaluatorforum.org/ — A coalition setting shared standards for credible third-party AI evaluation, through the Independent Evaluation Standard (AEF-1) and an Evaluation Transparency Letter. Not a job board, but its member list is effectively a map of the independent evals sector — AVERI, Collective Intelligence Project, Meridian Labs, METR, Princeton HAL, RAND, SecureBio, and Transluce, all listed above — and evals is one of the more accessible technical lanes for people arriving from security or testing work. Open to new member organizations.
• The Safety Apprentice — https://safetyapprentice.com/ — An AI safety job guide aimed at people working out how to enter the field.
• alphaXiv Researcher Profiles — https://www.alphaxiv.org/researchers — Researcher profiles built from published work. Useful for identifying who is actually working on a problem you care about, and who to name in an application.
• LessWrong — https://www.lesswrong.com/ — The forum where a large share of AI safety thinking happens first. Program announcements, research write-ups, and career roadmaps all land here.
• Alignment Forum — https://www.alignmentforum.org/ — The research-focused sibling of LessWrong, specifically for alignment work.
• EA Global — https://www.effectivealtruism.org/ea-global/ — Conferences run by the Centre for Effective Altruism, always with substantial AI safety content. EA Global itself is for people already in the community; the EAGx and EA Summit events are built for newcomers. Much of the funding on this page traces back to this movement, so it is worth knowing the room exists even if you do not identify with it.
➟ Next EA Global: San Francisco, March 19–21, 2027; then London, May 28–30, and Washington, DC, October 15–17
➟ For newcomers, taking applications now: EAGxBerlin (November 13–15), EAGxAustralasia (November 27–29), and EA Summits in Seattle (December 6) and Boston (December 12)
• AI Safety Nigeria — https://aisafetynigeria.com/ — Abuja, Nigeria — Research, training, and policy work on AI safety in Nigerian and African contexts, with virtual events and a paper reading group. Volunteer and partner routes in.
Written by people who did the hiring, or who made the transition themselves.
• Vlad Feinberg, How to Land a Frontier Lab Job
• Jack Morris, So you want to do AI research? It's true that no one really teaches you how.
• Neel Nanda, How To Become A Mechanistic Interpretability Researcher
• Neel Nanda, MATS 12.0 (Winter 2026-27) Info
• Mikhail Mironov, Roadmap Through AI Safety Programs for Early-Career Researchers
• Matt Beard, How to Get Into AI Safety in 3 Months
• Sophie Kim, Transitioning into AI safety? Design your own adventure
• Ben Lukszys, Dear builders, humanity needs your help
• Precious Oluwafemi Sani, AI Safety and Security From Scratch
Cohort dates, stipends, and application status were verified against each program's publicly available information on November 4, 2026. Feel free to use this page as a starting point in your search, but always confirm details with individual programs. Updates welcome: cleverhack at duck.com
|
|
|
|---|---|---|
|
|
|
The Great Frontier AI Lab Product War Is Just Beginning
|
Made with ♥ in Silicon Valley. Copyright © Joy Larkin 2002-2026.
contact
•••
speaking
•••
cleverhack.com
•••
llms.txt
•••