Global catastrophic risks
This board is still in beta. The reason below is our judgement — tell us if you think we have it wrong. Give feedback
In this list:At vetted organisations · 100Ways in · 38With leverage in the Netherlands · 4
Among the best in the world at what they do100 roles
This tier is about the solution itself. We are fairly confident these organisations are among the best in the world in their field — vetted by GiveWell, Animal Charity Evaluators, Founders Pledge or The Life You Can Save, or listed by 80,000 Hours, Probably Good, EA Opportunities or Animal Advocacy Careers. Somebody else has already judged that what they do works, so we list every role there that is open to someone in the Netherlands, including support and operations work. Expect fierce competition, and expect remote work or a move.
Project Tailwind: Frontier AI Evidence Generation
Coefficient GivingRemoteRemote
Those who decide on advanced AI in Washington currently rely largely on what the AI companies themselves say about their models.
AI safetyResearchPolicyMid levelProject Tailwind: AI Security & Verification
Coefficient GivingRemoteRemote
International agreements on powerful AI are only worth something if countries can check whether the other side is keeping to them, and the technology for that checking largely does not exist yet.
AI safetyEngineeringResearchSeniorProject Tailwind: AI Safety Public Goods
Coefficient GivingRemoteRemote
Setting up shared computing power, datasets and tested safety advice means working not on one piece of research but on the preconditions for many AI safety teams at once.
AI safetyPolicyResearchMid levelProject Tailwind: AI Safety & Governance Convenings
Coefficient GivingRemoteRemote
Work on safe and well-regulated AI is fragmented: researchers, policymakers and journalists who need each other often don't know each other.
AI safetyOperationsCommunications and outreachMid levelProject Tailwind: AI Red Teaming, Auditing, and Independent Assessment
Coefficient GivingRemoteRemote
AI companies currently assess the safety of their own models largely by themselves, and regulators and journalists can rarely test those claims.
AI safetyResearchSoftware engineeringSeniorProject Tailwind: AI Governance & Policy Research
Coefficient GivingRemoteRemote
A good way to test AI models only has to be worked out once: after that, other labs and regulators can adopt it without redoing the work themselves.
AI safetyResearchSoftware engineeringMid levelProduct Manager, Clinical AI
THINKMDRemoteRemote
As a doctor you help a few dozen patients on a busy day.
Global health and infectious diseaseManagementResearchMid levelHead of Red Teaming
SyntonyRemoteRemote
Whoever sets the testing method also determines which flaws testers find in AI systems before they reach users, and which they miss.
AI safetyResearchManagementSeniorOperations Lead
Odyssean InstituteRemoteRemote
At a small institute researching how risks such as unsafe AI could disrupt society, the director's time is the bottleneck: whoever takes over fundraising, recruitment and citizens' assemblies gives that time back to…
AI safetyOperationsManagementMid levelML Research Engineer
Compassion Aligned Machine LearningRemoteRemote
Whether a language model really behaves according to the values it is given during training is still hard to measure.
AI safetySoftware engineeringResearchMid levelSenior Media Manager, Communications
FAR AIRemoteRemote
FAR.AI investigates where advanced AI systems are vulnerable, but policymakers usually only read those findings in the newspaper.
AI safetyCommunications and outreachSeniorResearch Scientist
FutureSearchRemoteRemote
Anyone who wants to assess the risks advanced AI brings must first measure what those systems can really do and how reliable their forecasts are.
AI safetyResearchSoftware engineeringMid levelHead of Product
FutureSearchRemoteRemote
A forecast only helps if a policymaker trusts it and acts on it.
AI safetySoftware engineeringPolicySeniorDirector of Partnerships
FutureSearchRemoteRemote
Forecasts about AI only help if policymakers and labs trust and use them.
AI safetyFinanceManagementSeniorProgram Manager, Research
FutureSearchRemoteRemote
In a small research team you decide which lines of research start, continue or stop, and with that where all the researchers' time goes.
AI safetyManagementResearchSeniorSenior Researcher, AI Economic Preparedness
Windfall TrustRemoteRemote
No one knows how quickly AI will change jobs and tax revenues.
AI safetyResearchPolicyMid levelRegional Principal, AI Economic Preparedness
Windfall TrustRemoteRemote
How fast AI will shift work and incomes is uncertain, but how hard it hits depends on what governments have ready in advance.
AI safetyPolicyResearchSeniorConceptual Reasoning Researcher
Trajectory Labs, PBCRemoteRemote
Anyone who has AI models help think through hard questions, such as how to keep powerful AI safe and controllable, needs to know where those models sound convincing but reason badly.
AI safetyResearchSoftware engineeringMid levelPolicy and Communications Officer, AI in Europe
The School for Moral AmbitionBrusselsDutch employer, flexible
Europe's rules on AI are being decided now, and many policymakers and readers form their picture of the risks of advanced AI from a newspaper piece or a two-page briefing.
AI safetyPolicyCommunications and outreachMid levelTalent Acquisition Coordinator, Operations
Epoch AIRemoteRemote
Epoch AI publishes the figures on computing power and the growth of AI models that policymakers worldwide rely on when they draw up rules for AI.
AI safetyOperationsDataEntry levelTalent Acquisition Coordinator
Epoch AIRemoteRemote
Epoch AI grew from about 20 to 70 people in 2026, and policymakers worldwide use the figures on AI progress that the team publishes.
AI safetyOperationsEntry levelTransformative AI Fund
Effective Altruism FundsRemoteRemote
Whoever works here helps decide which new researchers and projects working on the risks of advanced AI receive funding.
AI safetyFinanceMid levelOperations & Programs Associate
AI Safety ConnectRemoteRemote
You won't be working on AI policy yourself.
AI safetyOperationsCommunications and outreachEntry levelResearch Scientist, Manipulation Evals
Apart ResearchRemote (EuropeRemote
A regulator can only act against AI that manipulates people once someone has shown how to measure it.
AI safetyResearchSoftware engineeringMid levelResearch Support
AI Alignment FoundationRemoteRemote
Many ideas for keeping AI systems safe, some drawn from neuroscience or physics, stay on paper because the person who came up with them has no team to test them.
AI safetyResearchSoftware engineeringMid levelSoftware Engineer, Product
Collective Intelligence ProjectRemoteRemote
Whether ordinary people's views count in how AI is developed depends partly on practical software: how you collect, organise and summarise thousands of responses without losing the nuance.
AI safetySoftware engineeringMid levelHead of Operations
QuorumRemoteRemote
Quorum does research and policy work to keep powerful AI systems controllable, and is growing from seven to at least twenty people.
AI safetyOperationsManagementSeniorML Researcher, Post-Training and Safety
Compassion Aligned Machine LearningRemoteRemote
If you teach a language model certain values, do they hold up when someone tries to mislead it or trains the model further?
AI safetyResearchSoftware engineeringMid levelAI Evaluation and Agent Systems Engineer
SyntonyRemoteRemote
Anyone who changes an AI model or an AI agent cannot tell, without good tests, whether it then behaves more safely or more riskily.
AI safetySoftware engineeringResearchMid levelHorizon Scanner, Scalable Alignment
Principles of IntelligenceRemoteRemote
You won't do the research yourself here; instead you work out which questions in AI safety – the field that aims to keep powerful AI systems under control – nobody is taking up yet.
AI safetyResearchPolicyMid levelAI Evaluation & Agent Systems Engineer
SyntonyRemoteRemote
Anyone who wants to know whether an AI system has become safer or riskier after an update needs a reliable way to measure it.
AI safetySoftware engineeringResearchMid levelResearch Grants
Center on Long-Term RiskRemoteRemote
The Center on Long-Term Risk funds research into how AI and other future technology could cause suffering on a large scale — a question few researchers are working on yet.
AI safetyFinanceMid levelRapid Grants
BlueDot ImpactRemoteRemote
This is not a vacancy but a rapid grant of up to $20,000 for anyone who wants to work on AI safety or biosecurity, reducing risks from pathogens.
AI safetyResearchEntry levelScientist, Plume Modeling and Forecasting
ReflectiveRemoteRemote
Whether it is ever wise to cool the earth with particles in the stratosphere depends on a question that is still open: how does such a plume behave in its first hours and days?
AI safetyResearchEngineeringMid levelProject Manager
QuorumRemoteRemote
AI safety — the field that aims to prevent powerful AI systems from causing serious harm — is still small, and researchers, policymakers and practitioners rarely end up in the same room.
AI safetyOperationsManagementMid levelHead of Special Projects, AI Safety
QuorumRemoteRemote
AI safety is the work of preventing powerful AI systems from causing serious harm, and the field is small: researchers, AI labs, policymakers and funders often work in isolation from one another.
AI safetyManagementResearchSeniorChief Operating Officer
Apart ResearchRemoteRemote
Apart Research trains new researchers and does its own research into AI safety: how to prevent powerful AI systems from causing serious harm.
AI safetyManagementOperationsExecutiveAI Industry Associate/Lead
Future of Life InstituteRemote (Europe)Remote
Countries such as Brazil, Canada and the EU member states are working on their own AI strategies, and most tacitly assume these must centre on ever larger, more autonomous systems.
AI safetyPolicyResearchMid levelResearch Mentor, AI Safety
PRISMRemoteRemote
You spend around 14 hours, spread over 16 weeks.
AI safetyResearchSeniorResearch Scientist, Manipulation Evals
Apart ResearchRemoteRemote
Regulators can only enforce rules for what they can measure, and whether an AI model manipulates people can still barely be measured reliably.
AI safetyResearchSoftware engineeringMid levelCommunications and Outreach, AI Safety
Cedar ResearchRemoteRemote
Research into the safety of AI systems only helps once policymakers, journalists and other researchers know about it, and small research groups rarely have someone who translates it for a broad audience.
AI safetyCommunications and outreachResearchEntry levelAI Enablement Manager
Epoch AIRemoteRemote
Policymakers and researchers around the world use Epoch AI's figures on computing power and the growth of AI to estimate how quickly this technology is developing.
AI safetyOperationsSoftware engineeringMid levelTechnical Specialist, AI Safety
Cedar ResearchRemoteRemote
Cedar Research publishes its code, methods and papers openly, so that other research teams build on them instead of working out the same things again.
AI safetySoftware engineeringResearchMid levelSpecial Projects Officer, Strategy
Ambitious Impact (AIM)RemoteRemote
Ambitious Impact (AIM) helps people found new charities, and a small leadership team decides which gaps in that landscape are tackled first.
Effective careersResearchOperationsMid levelSPAR Research Manager
KairosRemoteRemote
You guide mentors who each lead a research team in AI safety: research into how to stop powerful AI systems from causing serious harm.
AI safetyResearchManagementMid levelProgram Lead, Working Groups
Responsible AIxBio CollaborativeRemoteRemote
New AI models can help malicious actors design biological weapons, while the AI labs, biotech companies and policymakers who need to stop that rarely sit at the same table.
AI safetyManagementPolicySeniorOperations Lead
Talos NetworkRemoteRemote
Talos Network prepares people for policy work on AI in Europe: who gets control over this technology and what rules apply to it.
AI governance and powerOperationsFinanceSeniorHead of Operations
PRISM GlobalRemoteRemote
PRISM Global researches whether AI systems can experience anything, a question few people yet work on full time.
AI safetyOperationsManagementSeniorCo-Founder
Responsible AIxBio CollaborativeRemoteRemote
Companies building AI models for biological research do not yet have shared agreements to prevent misuse for biological weapons, and they rarely sit down together to discuss it.
Pandemics and biosecurityPolicyManagementExecutiveGeneralist, SPAR
KairosRemoteRemote
SPAR pairs people who want to start out in AI safety, the research into how to keep powerful AI systems safe, with experienced mentors.
AI safetyOperationsManagementMid levelChief of Staff
Responsible AIxBio Collaborative (R-ABC)RemoteRemote
R-ABC is a new non-profit that wants to use rules and agreements to stop anyone using AI to make a biological weapon or unleash a pandemic.
Pandemics and biosecurityOperationsManagementSeniorSenior AI Governance Grantmaker
Effective Institutions ProjectRemoteRemote
In this role you help decide where tens of millions of dollars go in a young field with few funders: the rules and institutions around powerful AI, and the question of who gets power over it.
AI governance and powerFinancePolicySeniorSPAR Research Manager
KairosRemoteRemote
You won't do research yourself, but you help the people who lead it: many of SPAR's 250 mentors are leading a team for the first time, and together they guide more than 800 newcomers to AI safety each round, the field…
AI safetyManagementResearchMid levelHead of Engineering, Red Team
FAR AIRemoteRemote
AI labs and governments let FAR.AI attack their newest models before they are released.
AI safetySoftware engineeringManagementSeniorRecruiter
KairosRemoteRemote
Kairos expects to double in size within six months, and each new hire runs programmes that hundreds of people pass through on their way to working on AI safety: the field that aims to prevent powerful AI systems from…
AI safetyOperationsCommunications and outreachMid levelResearch Engineer
IliadRemoteRemote
At Iliad, people research how to make powerful AI systems do what humans intend, and their pace often depends on how quickly rough test code becomes a reliable computation at scale.
AI safetySoftware engineeringResearchMid levelProof Engineer, Autoformalisation Engineer
IliadRemoteRemote
Mathematical research on the safety of AI systems depends on proofs that actually hold, and checking them by hand can cost researchers months.
AI safetySoftware engineeringResearchMid levelTechnical Grant Writer
AI Alignment FoundationRemoteRemote
AI safety researchers, who work out how to make powerful AI systems do what people intend, often write their own grant applications, and that comes out of their research time.
AI safetyCommunications and outreachResearchMid levelHead of Engineering (Red Team)
FAR AIRemote (International)Remote
FAR AI tests new AI models from OpenAI and Anthropic, among others, before they are released, and does the same for governments such as the European Commission: the team deliberately tries to make the systems go off…
AI safetySoftware engineeringManagementSeniorFunding for Programs and Events
Coefficient GivingRemoteRemote
Note: this is not a vacancy but an ongoing grant scheme from Coefficient Giving, formerly Open Philanthropy, one of the largest funders of work on major global risks.
Community buildingFinanceMid levelHead of Projects
Safe AI Netherlands (SAIN)AmsterdamOn site in the Netherlands
Many people take a course on AI safety — the field that tries to get powerful AI systems to do what we intend — and then get no further than reading.
AI safetyManagementResearchSeniorResearch Scientist
AI Verification and Evaluation Research Institute (AVERI)RemoteRemote
Lawmakers want powerful AI systems to be independently scrutinised, but how to carry out such an audit technically is still largely an open question.
AI safetyResearchSoftware engineeringMid levelCoordination and Accountability
Foresight InstituteRemoteRemote
More and more AI systems are working with each other and with people without any single party overseeing them, and how to make that safe is largely unknown.
AI safetyResearchSoftware engineeringMid levelProduct Manager
Survival and Flourishing Fund (SAF & SFF)RemoteRemote
The Survival and Flourishing Fund distributes money across dozens of organisations every year, and the people who decide where it goes work with software that you, as product manager, improve.
Global catastrophic risksOperationsManagementMid levelProduct Manager, Website
Epoch AIRemoteRemote
Policymakers, journalists and researchers who want to know how fast AI is progressing often get their figures from Epoch AI, and the website is where they find them.
AI governance and powerCommunications and outreachDataMid levelEvents Team Member, Facilitator
Safe AI NetherlandsNetherlandsOn site in the Netherlands
Still only a few people work on the question of how to keep powerful AI systems safe, and some of them got involved through a reading group or course like this one.
AI safetyCommunications and outreachEntry levelEducation Lead, Technical and Repository Owner
Safe AI NetherlandsNetherlandsOn site in the Netherlands
In the Netherlands relatively few people work on AI safety, the research into how to keep powerful AI systems safe and controllable.
AI safetyResearchSoftware engineeringSeniorRecruiter or Senior Recruiter
Epoch AIRemoteRemote
Epoch AI is a small research institute that tracks in numbers how fast AI is advancing, such as compute and scaling, and policymakers and researchers worldwide use those numbers.
AI safetyOperationsCommunications and outreachMid levelSenior Advisor, Advanced Guardrails for AI-Induced Incidents
Institute for Security and Technology (IST)RemoteRemote
Governments have barely any playbook for the moment an AI system unexpectedly slips out of human control.
AI safetyResearchPolicySeniorEvents Lead
Safe AI NetherlandsAmsterdamOn site in the Netherlands
AI safety is a small field.
AI safetyCommunications and outreachOperationsEntry levelEducation Course Facilitator
Safe AI NetherlandsNetherlandsOn site in the Netherlands
Few people yet work on AI safety — making sure increasingly powerful AI systems stay under control — and for many participants this course is the first step towards that work.
AI safetyCommunications and outreachResearchEntry levelQA or Testing Engineer
Survival and Flourishing Fund (SAF & SFF)RemoteRemote
The Survival and Flourishing Fund distributes millions each year among organisations working on the greatest risks to humanity, and its evaluators weigh applications in its own software.
Global catastrophic risksSoftware engineeringDataMid levelFull-Stack Software Engineer
Survival and Flourishing Fund (SAF & SFF)RemoteRemote
The Survival and Flourishing Fund distributes money across dozens of organisations through its own procedure, in which formulas combine the judgements of several evaluators into a single allocation.
Global catastrophic risksSoftware engineeringMid levelTeacher, Teaching Assistant, or Writer
IliadRemoteRemote
Only a few people work on the question of how to make powerful AI systems do what we intend, and the field grows mainly through new people learning it.
AI safetyResearchCommunications and outreachMid levelHead of Operations
SASHRemoteRemote
SASH researches AI safety: how to keep increasingly powerful AI systems reliable and under human oversight.
AI safetyOperationsManagementSeniorLead Research Scientist
Neo ResearchRemoteRemote
As research lead you decide which questions a whole team investigates, and which are left aside.
AI safetyResearchSoftware engineeringSeniorHead of Operations
Singapore AI Safety Hub (SASH)Remote (LondonRemote
In a young, fast-growing organisation, whoever sets up recruitment, salaries and security determines how many researchers and policy people can genuinely spend their time making the most powerful AI systems safer.
AI safetyOperationsManagementSeniorEducation Course Facilitator, Utrecht
Safe AI Netherlands (SAIN)UtrechtOn site in the Netherlands
AI safety – the work of keeping powerful AI systems safe and controllable – is a small field, and for many people a course like this is the first step into it.
AI safetyResearchCommunications and outreachEntry levelEvents Lead, Amsterdam
Safe AI Netherlands (SAIN)AmsterdamOn site in the Netherlands
AI safety is the field that studies how to keep increasingly powerful AI systems safe and controllable, and few people work on it compared with the questions still open.
AI safetyCommunications and outreachOperationsEntry levelEducation Lead
Safe AI Netherlands (SAIN)UtrechtOn site in the Netherlands
Relatively few people work on AI safety, making sure powerful AI systems do what we intend.
AI safetyCommunications and outreachResearchMid levelCommunity Manager, Amsterdam
Safe AI Netherlands (SAIN)AmsterdamOn site in the Netherlands
AI safety — making sure increasingly powerful AI systems stay safe and do what we intend — is a small field in the Netherlands, and people who want to get into it often don't know where to start.
AI safetyCommunications and outreachOperationsEntry levelExpression of Interest in a Future Role
Sentinel BioUKDutch employer, flexible
In this role you don't do research yourself, but you help decide which research gets done.
Pandemics and biosecurityFinancePolicyMid levelResearch Scientist, Member of Technical Staff
AI DigestRemoteRemote
Governments and safety teams are already making decisions about AI systems that carry out tasks on their own, while what those systems do in practice is still poorly understood.
AI safetyResearchSoftware engineeringMid levelShortform Video Creator
AI DigestRemoteRemote
Most people form their picture of AI not from research reports but from what turns up on their phone.
AI safetyCommunications and outreachSoftware engineeringEntry levelResearch Scientist
AI DigestRemoteRemote
Most AI tests measure what a model does in a single short task.
AI safetyResearchSoftware engineeringMid levelEngineer
AI DigestRemoteRemote
AI Village shows in a public livestream what the latest AI agents do when they pursue a goal on their own for hours, including where they get stuck or do something unexpected.
AI safetySoftware engineeringEntry levelDesign Engineer
AI DigestRemoteRemote
In the AI Village, AI agents work independently on tasks for long stretches, and what happens there shows what these systems can already do and where they go off the rails.
AI safetySoftware engineeringCommunications and outreachMid levelExpression of Interest: Special Projects Associate
Epoch AIRemoteRemote
Epoch AI's figures on compute and scaling are cited by policymakers and researchers worldwide; you don't do that research yourself, but you make sure a small team can publish more by coordinating experts, licensing…
AI safetyOperationsResearchEntry levelVice President, The AI Access Initiative
Evidence ActionRemoteRemote
Evidence Action only takes a programme to millions of people once the evidence is there.
Global health and infectious diseaseManagementPolicySeniorProgram Lead, Career Transition Grants
BlueDot ImpactRemoteRemote
Anyone who wants to move into AI safety, the research into how to keep increasingly powerful AI systems safe and controllable, often has to retrain for months without an income, and that holds suitable people back.
AI safetyFinanceManagementMid levelAI Offense-Defense Dynamics Lead Researcher
Future of Life FoundationRemoteRemote
Whether a new AI system mainly strengthens attackers or defenders probably depends on properties you can measure in advance, such as how broadly it can be used and how easily anyone can get access to it.
AI safetyResearchPolicySeniorSenior Technical Specialist, AI Risk Assessment
Future of Life FoundationRemoteRemote
Whoever decides how the risks of advanced AI are measured helps decide what governments, researchers and funders pay attention to.
AI safetyResearchPolicySeniorResearch Engineer, Novel AI Platforms for Multiscale Alignment
Future of Life FoundationRemoteRemote
Many ideas for getting AI systems to work together safely still exist mostly on paper, because there is no good environment to test them in.
AI safetySoftware engineeringResearchMid levelFunding for Addressing Risks From Transformative AI
Coefficient GivingRemoteRemote
This is less a job than a rolling grant scheme: you apply to Coefficient Giving (formerly Open Philanthropy, one of the largest philanthropic funders in this area) for money for a course, fellowship, event or podcast…
AI safetyFinanceMid levelSenior Researcher, Civil Safety Sciences
Future of Life FoundationRemoteRemote
For many hazards that AI makes possible, such as synthetic microbes or new ways of spreading pathogens, there is as yet no field of study and no historical precedent to lean on.
AI safetyResearchPolicySeniorGeneral Expression of Interest (EOI)
KairosRemoteRemote
Kairos does no AI safety research itself, but trains the people who will: more than 1,200 participants in two years, through fellowships and student groups.
AI safetyManagementCommunications and outreachEntry levelHead of Talent Operations
KairosRemoteRemote
Kairos trained more than 1,200 people in two years, but nobody systematically matches early-career researchers with open positions at AI safety organisations.
AI safetyOperationsManagementSeniorSPAR Research Manager
KairosRemoteRemote
Each round, SPAR pairs more than 400 early-career researchers with 130 mentors at places such as Anthropic, RAND and the UK AI Security Institute.
AI safetyManagementOperationsMid levelGeneralist (Talent Ops)
KairosRemoteRemote
Kairos trained more than 1,200 people for work on AI safety in two years, and is growing from 7 to around 20 staff.
AI safetyManagementOperationsMid levelAI Safety Argumentation Platform Research Engineer
Future of Life FoundationRemoteRemote
Anyone who has to make decisions about the risks of AI currently gets mostly separate reports and opinions that contradict each other.
AI safetySoftware engineeringResearchMid level
Ways in38 openings
This band is about where it leads. Internships, traineeships and paid programmes — fellowships, residencies, bootcamps — are rarely where you change the most, but they are often how people get into the teams where that happens. An internship with the directorate writing the national AI strategy is how people end up writing it later. Some programmes are held abroad. We list those when they are paid and last six months or less, because then you do not have to move for them.
Project Tailwind: Talent Acceleration
Coefficient GivingRemoteRemote
AI safety, the work of keeping increasingly powerful AI systems safe, seems above all to be short of experienced people who can set up and lead new organisations.
ProgrammeAI safetyManagementResearchMid levelTalos Incubator Europe
Talos NetworkRemoteRemote
You won't be joining an existing organisation here; you'll be founding one: a team that helps policymakers implement the European AI rules and limit the risks of advanced AI, and that hires people and keeps going after…
Programme · 26 weeksAI safetyResearchPolicyEntry levelTalos Fellowship
Talos NetworkRemote (Europe) (Europe (excluding UK)Dutch employer, flexible
Europe is now writing the rules for the most powerful AI systems, but few people understand both the technology and the policy work in Brussels.
Programme · 9 weeksAI safetyPolicyResearchEntry levelTalos Incubator Europe: found an AI governance organisation
Talos NetworkRemote (Europe)Dutch employer, flexible
You are not building a career within an existing organisation here.
Programme · 26 weeksAI safetyPolicyResearchEntry levelInternational Programme on AI Evaluation: Capabilities & Safety
AI Evaluation ProgrammeRemoteRemote
There is no established method yet for testing whether an advanced AI system is safe enough: governments and AI companies are only now looking for those standards.
Programme · 22 weeksAI safetyResearchPolicyEntry levelSummer Research Fellowship, EU Law
Institute for Law & AICambridgeTemporarily abroad
The EU AI Act has been adopted, but how strict it turns out to be in practice depends on implementing rules and guidelines that are still being written.
Programme · 10 weeksAI safetyLegalPolicyEntry levelStage bij de Directie Integratie Europa
Ministerie van Buitenlandse ZakenRijnstraat 8 te Den HaagOn site in the Netherlands
This is where the ministries agree each week on the position the Netherlands takes in Brussels, and the EU decisions that position feeds into set rules for 450 million people and a large share of European development…
Democratic institutionsPolicyResearchEntry levelDutch requiredSummer Research Fellowship, EU Law
Institute for Law & AI (LawAI)CambridgeTemporarily abroad
The EU AI Act is in place, but how regulators and courts will interpret it is still largely open.
Programme · 10 weeksAI safetyLegalResearchEntry levelResearch Resident, US
AnthropicRemoteTemporarily abroad
AI safety is a small field: a method for seeing what happens inside a language model, devised by a handful of researchers, is soon used by other labs too.
Programme · 26 weeksAI safetyResearchSoftware engineeringMid levelResearch Resident, UK
AnthropicUKTemporarily abroad
Here, research into what happens inside AI models and when they go off the rails goes straight to the people who train Claude, a system used by millions of people.
Programme · 26 weeksAI safetyResearchSoftware engineeringMid levelDC Summer Fellowship (2027)
Centre for the Governance of AI (GovAI)WashingtonTemporarily abroad
How the US regulates powerful AI is largely decided in Washington by a small group of policymakers, who seek advice from think tanks such as GovAI.
Programme · 13 weeksAI safetyResearchPolicyEntry levelUK Summer Fellowship (2027), Research Track
Centre for the Governance of AI (GovAI)LondonTemporarily abroad
GovAI is a London research institute that helps policymakers at governments and AI companies work out how to steer advanced AI in the right direction.
Programme · 13 weeksAI safetyResearchPolicyEntry levelUK Summer Fellowship (2027), Applied Track
Centre for the Governance of AI (GovAI)LondonTemporarily abroad
According to GovAI, the scarce people in work on rules and oversight for advanced AI are not mainly researchers, but people who organise, communicate and lead policy processes.
Programme · 13 weeksAI safetyPolicyOperationsEntry levelSeed Grants
BlueDot ImpactRemoteRemote
This is not a job but start-up money: 20,000 to 200,000 dollars to spend six to twelve months setting up your own project in AI safety or biosecurity (preventing disasters involving pathogens), areas where few people…
Programme · 26 weeksAI safetyResearchSoftware engineeringEntry levelEmergent Ventures Fellowship
Mercatus Center at George Mason UniversityRemoteRemote
This is not a job but a grant: you submit an idea that does not exist yet, and if Emergent Ventures picks it, you get money to set it up, often before other funders have looked at it.
ProgrammeEffective givingFinanceEntry levelTeacher, Technical AI Safety/Governance & Strategy
ML4GoodRemoteRemote
Still relatively few people work on keeping powerful AI systems safe, and the field grows mainly through new people learning the trade.
Programme · 2 weeksAI safetyResearchCommunications and outreachMid levelFellowship, Epidemiology and Public Health Microbiology
European Centre for Disease Prevention and Control (ECDC)RemoteRemote
Whether an outbreak of an infectious disease stays local often depends on how quickly someone on the ground works out where it came from and how it is spreading.
Programme · 104 weeksPandemics and biosecurityResearchPolicyEntry levelExecutive Catalyst: Ops & Strategy
SteadRiseRemoteRemote
Organisations trying to reduce serious risks from advanced AI are growing fast, but they have few people who have built an organisation before.
Programme · 12 weeksAI safetyManagementOperationsMid levelUndergraduate Intern, AI Safety Research and Engineering
SyntonyRemoteRemote
As an intern you won't solve much yourself: the value lies in what you take away.
ProgrammeAI safetyResearchSoftware engineeringEntry levelHumanitas Fellowship
SyntonyRemoteRemote
On how AI can be misused, and how to keep telling real from fake now that synthetic images are increasingly convincing, there is still little careful research for policymakers and lawyers to build on.
Programme · 35 weeksAI safetyResearchPolicyMid levelCareer Transition Grants, AI Safety and Biosecurity
BlueDot ImpactRemoteRemote
This is not a job but a grant: BlueDot Impact gives between 20,000 and 200,000 dollars so that you can move full-time into work on AI safety or biosecurity, rather than trying to do so alongside your current job.
ProgrammeAI safetyResearchEntry levelResearch Mentorship, Alignment Research
Supervised Program for Alignment Research (SPAR)RemoteRemote
Only a few experienced researchers worldwide work on the question of how to make powerful AI systems do what we intend — AI safety — and their guidance is what beginners lack most.
ProgrammeAI safetyResearchEntry levelFacilitator, Research Fellowship
MercialRemoteRemote
Anyone starting out in AI safety research, meaning research into how to keep powerful AI systems safe, rarely gets good feedback on a first piece of work, even though that feedback often decides whether they carry on.
Programme · 4 weeksAI safetyResearchEntry levelFunding, AI Safety
The Global Technology Risk FoundationRemoteRemote
AI safety, the research into how to keep powerful AI systems reliable and controllable, is a young field with few people in it.
ProgrammeAI safetyResearchEntry levelResearch Fellow, AI Interpretability
GoodfireUSATemporarily abroad
Large AI models are trained by trial and adjustment, without anyone properly understanding why they do what they do.
Programme · 13 weeksAI safetyResearchSoftware engineeringEntry levelAlignment Fellow
Arcadia ImpactLondonTemporarily abroad
Large AI companies know a lot about the techniques they use to make their models do what is intended, but those on the outside see little of it.
Programme · 13 weeksAI safetyResearchSoftware engineeringEntry levelScience and Technology Policy Fellow, Legislative Branch Cohort in AI
American Association for the Advancement of Science (AAAS)WashingtonTemporarily abroad
Members of the US Congress decide on AI legislation, yet their offices rarely have anyone who understands the technology itself.
Programme · 2 weeksAI safetyPolicyResearchEntry levelDutch nationality requiredRequest for Proposals: Project Tailwind
Coefficient GivingRemoteRemote
This is not a vacancy but a funding round: Coefficient Giving (formerly Open Philanthropy) is looking for people to set up a new organisation or research group in AI safety, the work meant to ensure powerful AI systems…
ProgrammeAI safetyManagementResearchEntry levelIliad Intensive
IliadUKTemporarily abroad
Research into how to keep powerful AI systems safe and predictable (AI safety) is a young field, and only a few people are working on its basic mathematical questions.
Programme · 4 weeksAI safetyResearchSoftware engineeringEntry levelIliad Fellowship
IliadUKTemporarily abroad
AI safety, the study of how to make powerful AI systems reliably do what people intend, is a young field with few mathematicians.
Programme · 26 weeksAI safetyResearchSoftware engineeringEntry levelSummer Research Fellowship (Legal Frontiers)
Institute for Law & AIRemoteRemote
Legislators in the US, the EU and the UK are writing rules for AI now, but how you make an AI system follow the law is a question few lawyers are working on yet.
Programme · 10 weeksAI safetyLegalResearchEntry levelSummer Fellowship (US)
Institute for Law & AIUSATemporarily abroad
The law around AI is being written now, while only a few lawyers are engaging with it seriously.
Programme · 12 weeksAI safetyResearchPolicyEntry levelSummer Fellowship
Institute for Law & AI (LawAI)WashingtonTemporarily abroad
Few lawyers know both the law and AI well, while governments are drawing up the rules for that technology right now.
Programme · 12 weeksAI safetyLegalResearchEntry levelReporting Grants & Fellowships
Pulitzer CenterRemoteRemote
Some big problems almost nobody writes about, and what stays unknown rarely gets money or policy.
ProgrammeEffective givingCommunications and outreachResearchEntry levelAI Safety Residential Community
Centre for Enabling EA Learning & Research (CEEALAR)UKTemporarily abroad
CEEALAR pays for your room and board in Blackpool (England), so that you can work full-time on AI safety for six weeks to a year: preventing powerful AI systems from causing serious harm.
Programme · 26 weeksAI safetyResearchEntry levelAIXI Labs Research Fellowship
AIXI LabsRemoteRemote
A lot of AI safety research is mathematically rigorous but too narrow, or practically usable but too opaque to base hard safety claims on.
Programme · 13 weeksAI safetyResearchSoftware engineeringEntry levelResearch Fellows Program
TimaeusRemoteRemote
Tenured mathematicians and physicists rarely work on the question of what happens inside an AI model during training, even though their fields, such as algebraic geometry or statistical physics, seem to have exactly…
Programme · 52 weeksAI safetyResearchSoftware engineeringMid levelExpressions of Interest
Mirror Biology Dialogues FundRemoteRemote
Mirror life (organisms built from mirrored versions of the molecules of all natural life) does not exist yet, but 38 scientists warned in Science in December 2024 that a mirror bacterium could cause unprecedented harm…
ProgrammePandemics and biosecurityResearchEntry level
Roles with leverage, here in the Netherlands4 roles
This tier is about leverage. Most of these employers — ministries, regulators, universities, banks — would not say they work on one of the world’s largest problems, and nobody has vetted them as a whole. What we look for is different: roles that let you point more at the problem than your own hours. That might be money, or rules that bind a whole sector, or the work of dozens of other people, or an idea that is free to copy afterwards. A credit analyst at an agricultural lender can matter more than a programme officer at a charity. That judgement is ours, made role by role, and we always write down the reason. Competition here is usually a good deal lighter.
Senior inspecteur Markttoezicht Digitale Veiligheid
Rijksinspectie Digitale InfrastructuurAmersfoortOn site in the Netherlands
From 2027 the EU Cyber Resilience Act requires manufacturers to make software and hardware secure by design, and the RDI, the Dutch digital infrastructure inspectorate, is building up supervision of it now.
AI safetyPolicyInformation securitySeniorDutch requiredCoordinerend managementondersteuner PPG
RIVMBilthovenDutch employer, flexible
In a future pandemic, the crisis organisation of RIVM, the Dutch national public health institute, has to scale up within days, and whether its infectious disease experts then get to do their own work depends partly on…
Pandemics and biosecurityOperationsManagementMid levelDutch requiredSenior Ontwikkelaar / Beheerder Datastream
Ministerie van DefensieUtrechtOn site in the Netherlands
Your leverage lies in the team, not in your own code: by coaching developers and removing obstacles, you keep the data platform the Koninklijke Marechaussee, the Dutch military police and border guard, relies on for…
AI safetySoftware engineeringManagementSeniorDutch requiredAfdelingshoofd Informatievoorziening
RIVMBilthovenDutch employer, flexible
When a new pathogen emerges, the RIVM, the Dutch national public health institute, can only respond as fast as its information systems allow.
Pandemics and biosecurityManagementOperationsSeniorDutch required
New roles in your inbox
On the first Tuesday of every month: new jobs and programmes from this board, plus events in the next six weeks and the best of the EA Forum.
What do you make of this?
The board is still in beta, so this is exactly the moment to say what you think. A vacancy we have missed, a judgement you do not share, something that does not work — all of it helps.
Want to understand this first?
Start with our free course: about 100 minutes in six short units, at your own pace, with no sign-up. To go further, apply for the EA Fellowship: five weeks of reading and discussion in a small group. Also free, but selective.