Home
Start the quick course
Career guideAI safety

Work on AI safety from the Netherlands

AI is getting more capable fast. You can help it go well from here: in government, in research, or by starting something the field still lacks.

Browse AI safety jobs Book a call

About 15 minutes to read. Written in 2023 from interviews with people in the field, updated November 2026.

At a glance

AI safety in the Netherlands

  1. The biggest gap is founders
  2. You don't need an AI background
  3. Governance has more room for newcomers
  4. The Netherlands has real leverage
  5. Start with a free course near you

We've helped people into AI safety since we organised our first AI safety retreat in 2022.

Part 1: Why AI safety, why here

AI is moving fast, and the Netherlands has leverage

AI systems are getting more capable quickly. METR found that the length of tasks AI agents can complete on their own has been doubling roughly every seven months. That could do enormous good, but it also brings serious risks: AI used for cyberattacks or to help build biological weapons, power concentrated in very few hands, and systems that act in ways their developers can't reliably oversee or control.

In 2023, hundreds of AI researchers and industry leaders signed a statement that reducing the risk of extinction from AI should be a global priority. AI safety is the work of reducing these risks, especially the most severe ones, through technical research, governance and building the field. The Center for AI Safety has a readable overview of the risks.

Where the Netherlands fits

The rules are made in Brussels
The EU AI Act has been law since August 2024. Since August 2025 it has also covered general-purpose AI models, supervised by the European Commission's AI Office. Because any company selling in the EU has to follow EU rules, they shape AI far beyond Europe.
The Netherlands decides how they're enforced here
The government has chosen to share supervision among existing regulators, coordinated by the Autoriteit Persoonsgegevens (AP) and the Rijksinspectie Digitale Infrastructuur (RDI). The law that sets this up was still being finalised in late 2026, and people who understand AI are scarce in all of these bodies.
A middle power that's starting to act
The Netherlands is working with other middle powers on AI safety and exploring its own AI safety institute (see part 4).
ASML
The world's most advanced AI chips can't be made without ASML's machines. That makes the Netherlands an important place for compute governance.
Strong universities
The University of Amsterdam is one of Europe's leading places for machine learning.
A growing community
Safe AI Netherlands, the Existential Risk Observatory and Catalyze are all based here, and Dutch researchers co-founded the research organisation Resolution (formerly Timaeus). Read Jesse Hoogland's story.

About this guide

It gives you the Dutch specifics. For the wider picture, read 80,000 Hours on AI careers and technical AI safety research. It's for students and people early in their careers, and for experienced professionals, whom the field especially needs.

Whatever path you consider, how well it suits you matters as much as the path itself. Our career guide shows how to test your fit cheaply.

Part 2: The biggest gap

The field needs founders

Ask the people who fund AI safety what the field needs most, and many will say founders. Coefficient Giving, one of the largest funders in the field, says many critical problems have no one working on them, and launched Project Tailwind to find people to start new organisations. Catalyze Impact, which incubates AI safety organisations, says funders have set aside enough to build nonprofits worth $100 million or more.

The money is there. What's missing is people who can turn an idea into an organisation.

Who should consider it

Funders are especially interested in people with strong technical ability, experience building complex or security-critical systems, machine learning research expertise, experience in auditing or red-teaming, a track record of founding or scaling companies, nonprofits or programmes, or a talent for communicating with technical and non-technical audiences. You don't need all of these, and you don't need to be a researcher.

Where to get support

Project Tailwind
From Coefficient Giving: pre-seed funding of $200,000 to $2 million to develop an idea over 9 to 18 months, then seed funding of $2 million to $20 million to launch. It funds work outside the US and pays competitive salaries.
Catalyze Impact
An incubator for experienced founders, started in the Netherlands. It helped launch Resolution.
Talos Incubator Europe
Helps you find a co-founder and build a new AI governance organisation in Europe on a monthly stipend, ending with a pitch for seed funding.
BlueDot Impact seed grants
Funding for the first 6 to 12 months of a new project or organisation.
Generator
A paid three-month residency, run by Kairos, where you pitch, build and ship projects with mentors from across the field.

Before you start

Several organisations have already started here, from Catalyze and Resolution to Safe AI Netherlands. And there are still clear gaps, especially in European AI governance and in making the AI Act work in practice.

Talk to people already working on the problem. They may be better placed to do it, or become your first supporters. And talk to us: we can introduce you to founders in the Dutch AI safety community. Our founding guide covers testing an idea and setting up in the Netherlands.

Part 3: Choose a path

Three ways in

Not everyone should found something. These are the other main ways to help, and the rest of this guide covers each one.

AI governance

Shaping the laws, institutions and norms around AI, in government, regulators, standards bodies and think tanks. See part 4.

Good fit if you like writing, people and politics, and can wait years for results.

Technical AI safety

Research to make AI systems behave as intended and stay under human oversight. See part 5.

Good fit if you have, or can build, strong maths and machine learning skills.

Building the field

Courses, communities and programmes that help more people do this work. See part 6.

Good fit if you're good with people and projects, and don't want to be a researcher.

AI safety organisations also need people in operations, management, communications and finance, and many of those roles need no technical background.

Part 4: AI governance

Shape the rules

AI governance is about the norms, laws, institutions and politics that shape how AI is developed and used. Much of it happens in government, but companies, standards bodies, think tanks and civil society all play a part. This part focuses mainly on government policy.

In short

  • The US and Brussels matter most, but middle powers like the Netherlands are starting to act together on AI safety, so roles in the Dutch government can matter more than you might think.
  • Career capital matters more than people expect. In policy, your reputation and credibility are much of what you have to offer.
  • Joining a political party is an accessible way to contribute and to test your fit for policy work.
  • It's often easier to learn the technical side first and pick up governance skills later.

The case for working in the Dutch government

The US and China dominate AI development, but they aren't the only countries that matter. Middle powers such as the Netherlands, Canada, South Korea and Singapore can shape AI governance where the superpowers won't: by agreeing safety standards together, by building the expertise to test AI systems themselves, and, some argue, by making access to their combined markets depend on meeting those standards.

Finland shows what a small country can do. In September 2026, Finland's president and Norway's prime minister launched A Call for Control of Frontier AI Models. It asks for mandatory testing of frontier models before deployment, independent evaluation, shared reporting of serious incidents, and an international body that can set standards and check they're met. More than 20 leaders and the European Commission backed it, including Prime Minister Rob Jetten. Neither the US nor China signed.

The Netherlands is starting to move too. Its International AI Strategy (July 2026) commits to working with like-minded middle powers, exploring international governance of compute, seeking to join the international network of AI safety institutes, and exploring whether to set up a Dutch AI safety institute. In November 2026 the cabinet said it was studying an institute modelled on the UK's AI Security Institute, with advice due in December.

The next few years will bring new teams, mandates and diplomatic positions, and very few Dutch civil servants know much about the risks from frontier AI. If you do, a role at Foreign Affairs, Economic Affairs or a future Dutch AI safety institute could give you real influence.

What to study

A technical background, in machine learning, chips or computer science, can be very valuable. Most politicians and civil servants are generalists who rely on experts for technology they don't fully understand. If you can explain it clearly, you can be very useful.

You don't need to be an AI expert, though. Politics, public administration, law, international relations and economics are all good routes. A strong combination is a technical bachelor's followed by a master's in policy. You can also learn a lot on your own, for example through a free course (see part 6).

For jobs in the Netherlands, it doesn't matter much which Dutch university you go to. If you're aiming for Brussels, graduates from the College of Europe, Sciences Po and the London School of Economics have an edge.

Master's programmes in the Netherlands (alphabetical, not a ranking)

Where to work in the Netherlands

Your first policy jobs are mostly about building career capital: a network, credibility and an understanding of how government works. Much of that carries over between roles, so a policy job that isn't about AI can still be a good first step.

Ministries
The Ministry of Foreign Affairs leads the international AI strategy and AI diplomacy. The Ministry of Economic Affairs and Climate Policy (EZK) and the Ministry of the Interior and Kingdom Relations (BZK) lead on AI policy, including the AI Act.
Regulators
The Autoriteit Persoonsgegevens coordinates supervision of algorithms and AI, and the RDI the technical side.
Standards
NEN represents the Netherlands in the European committees writing the technical standards that companies will use to show they follow the AI Act. It's niche work with outsized influence.
Research and advice
TNO, the Rathenau Instituut and the WRR advise government on technology.
Traineeships
The Rijks I-traineeship and the other government traineeships are good ways in for graduates. Our policy guide has more.

Our jobs board lists AI safety roles you can do from the Netherlands.

Brussels and beyond

The European Commission
Its AI Office oversees the most capable AI models. The Blue Book traineeship is the usual way in for graduates.
Talos Fellowship
The Talos Fellowship combines a course in EU AI policy with a placement in Brussels. EU citizens are strongly preferred.
The UK and US
The UK's AI Security Institute hires technical and policy staff. For US policy careers, see Emerging Tech Policy Careers.
Remote roles
The 80,000 Hours job board lists many.

Government or elsewhere?

Besides government itself, you could work at a think tank, a research organisation or an advocacy group. Some things to weigh up:

For government

Network and credibility
Government builds both better than most places.
A few people can matter a lot
One or two well-placed people can change a decision, and few think tank reports are read by the officials who decide.
Trust comes quickly
Many civil servants know little about AI, so even a modest understanding earns trust.

Against government

It's slow
Governments often prefer to wait and see, and are reluctant to take large risks seriously.
Long feedback loops
It's hard to know whether you're helping.
Hierarchy and politics
The culture doesn't suit everyone. And you work for one government, while a think tank's research can reach several.

Compute governance

Compute governance is about who gets access to the computing power needed to train the most capable AI systems. Those systems are trained on huge amounts of specialised hardware, which is easy to detect and comes from a very concentrated supply chain. That makes it one of the more practical levers for governing AI.

ASML is the only company that makes the EUV machines needed for the most advanced chips, so the Netherlands is a natural place for this work, and few Dutch or European civil servants understand chips or their supply chains in depth. A technical degree such as computer or electrical engineering helps a lot, but a good grasp of how chips are made and traded is also valuable. If you want to explore this, talk to us.

Join a political party

Joining a political party is one of the easiest ways to have an effect now and build career capital. It can take surprisingly little effort to influence a party's programme. You'll also learn to explain AI risks to people who know little about them, and to talk constructively with people who disagree. And you'll find out whether political work suits you.

Common mistakes

Not considering all your options
Many people only look at jobs in their home country or at a few organisations. Apply broadly.
Writing policy opinions online instead of building career capital
Blog and forum posts do little for your reputation in policy. Volunteering for an advocacy organisation or working in government does more.
Expecting impact straight away
Your first few years are mostly about learning and credibility. Optimise for the long run.
Chasing status
A less glamorous role can teach you more and lead somewhere better.
Part 5: Technical AI safety

Make AI systems safe

Technical AI safety is research into making AI systems behave as intended and stay under meaningful human oversight. It happens at AI companies, nonprofits and universities, and among independent researchers. Research scientists come up with ideas, design experiments and analyse results; research engineers build, scale and run them. The line is blurry, but knowing which suits you helps you plan which skills to build.

In short

  • For a master's in AI at the University of Amsterdam, a bachelor's in maths, physics or computer science may prepare you better than a bachelor's in AI.
  • The Netherlands isn't a hub for technical safety work, but it's a very good place to study and to build skills.
  • Competition for full-time safety research jobs is intense. Being excellent in a niche can beat being average in machine learning.
  • Visible work, like code, papers and write-ups, counts for a lot.

What to study

Machine learning and AI degrees are the most direct route. Data science, applied maths, physics, logic, philosophy, cognitive science and neuroscience can also produce strong safety researchers.

Surprisingly, a bachelor's in AI isn't always the best start. It may not give you the mathematical depth that demanding master's programmes expect, and much of it is repeated in the master's. Maths, physics or computer science often give a stronger foundation. Check what your target master's requires before you choose.

The University of Amsterdam is the clear leader in deep learning in the Netherlands, and its MSc Artificial Intelligence is our top recommendation. It gives you a strong technical grounding, with specialisations such as natural language processing, reinforcement learning and computer vision. Other universities may offer less depth but more freedom to choose courses.

Other master's programmes (alphabetical)

Should you do a PhD?

Topic
Pick something close to alignment if you can. If not, choose a related area such as large language models or reinforcement learning.
Supervisor
Research skills transfer between topics, so a supervisor you work well with can matter more than the exact topic.
Credentials
Some research scientist roles ask for a PhD. Engineering roles usually don't.
Skills and stability
A PhD is essentially four years of entry-level research, with more security than many industry research jobs.

Where to work

Full-time technical safety jobs are scarce, even for very skilled people, and few are in the Netherlands. Some options:

Remote roles at safety organisations
Mostly based in the US and UK. Look at the 80,000 Hours job board, AISafety.com and our jobs board.
Academia
The Netherlands has more openings here. Look at the AI labs at the University of Amsterdam, TU Delft and Utrecht University, and at the ELLIS unit in Amsterdam.
An engineering job in industry
This teaches you to build robust software and machine learning systems quickly, and safety organisations value that experience. It also gives you savings for riskier moves later. The downsides: your skills may not be very safety-relevant, it's hard to keep up with the field next to a full-time job, and a comfortable salary can make it harder to switch. 80,000 Hours has a guide to moving from software engineering into AI safety.

Independent research

You can also build a research career through self-study and independent projects, in your free time or with a grant. The Long-Term Future Fund and BlueDot Impact's career transition grants fund people doing this. It makes sense when nobody is working on the direction you find most promising, or to build skills between jobs. It's hard, though: there's no structure, so you need a lot of self-discipline. If your idea is bigger than one person, see part 2.

Common mistakes

Not making your work visible
Put projects on GitHub, write up what you find, and have a simple personal website.
Expecting impact straight away
Excellence takes years, and most people do their best work later in their careers.
Reading without doing
Reading papers and forums is useful, but you learn more by writing, discussing and asking researchers questions. Share rough ideas.
Chasing online status
Recognition on forums is easier to get than skill in maths and machine learning, and worth much less.
Having only one skill
Mainstream safety research only hires the very top. Experience in managing projects or people can help you stand out.
Neglecting your health
Even if you think powerful AI is coming soon, find a pace you can keep up.
Having no savings
Hiring in this field is slow. A financial buffer lets you wait for the right role instead of taking the first one.
Part 6: Programmes

Programmes that help you move in

Building the field is a career path of its own: helping more talented people understand AI risk and find their way into the work. The programmes below are the result, and they're a good way in.

Start in the Netherlands

The easiest place to start is Safe AI Netherlands (SAIN). It grew out of the AI Safety Initiative Groningen and became a national foundation in 2026. It runs:

Free in-person courses
On AI safety, technical AI safety and AI governance. No background needed.
Discussion groups, talks and hackathons
Open to anyone, in each chapter's city.
A project hub
Matches emerging researchers with experienced supervisors.
Chapters
In Amsterdam, Utrecht and Groningen, with Eindhoven, Delft and The Hague next.

More programmes

Courses and research

For students

  • Pathfinder FellowshipMentorship and funding for students who run AI safety or AI policy groups at their university. Run by Kairos.
  • Global Challenges ProjectFree three-day residential workshops on AI safety and biosecurity with experts, with travel support. Run by Kairos.

For experienced professionals

  • SuccessifFree, personalised career advising for moving into AI risk reduction, and an AI policy strategy fellowship.
  • SteadRiseHelps experienced people into senior AI safety roles, including three months of free one-to-one support for leaders. Its CEO, Roy Hagemann, did our EA Fellowship.
  • GeneratorA paid residency for people who want to work in AI safety but not as researchers. Run by Kairos.

Community

Next steps

What to do next

Take a course
Start with our quick course, then join a free course from SAIN or BlueDot, or a SPAR research project.
Start something
If you have an idea for a project or organisation, look at Project Tailwind and the other support in part 2.
Talk to us
Book a virtual coffee and we'll help you work out your next step and introduce you to people.
Use your thesis
Effective Thesis offers free coaching if you want your thesis to count.
Stay up to date
Read the AI Safety Newsletter from the Center for AI Safety and, for EU policy, the EU AI Act Newsletter.

Funding

Project Tailwind
Coefficient Giving's funding for people starting ambitious new AI safety organisations.
BlueDot Impact grants
Seed grants for the first 6 to 12 months of a new project, career transition grants to spend a period working full-time on moving into AI safety, and rapid grants. BlueDot took over career transition funding from Coefficient Giving (formerly Open Philanthropy) in 2026.
Long-Term Future Fund
Grants for upskilling, independent research and new projects.
Future of Life Institute
Grants, including PhD and postdoctoral fellowships.
“The huge network I've gotten in contact with through EA has been amazing. I've made this pivot much, much faster than I think I could have accomplished in other comparable fields.”
Jesse Hoogland
Co-founder of Resolution, an AI safety non-profit

Want to talk it through?

A call with our team is free. We'll help you work out your next step and introduce you to people working on AI safety in the Netherlands.

Book a call Browse AI safety jobs

We wrote the first version of this guide in 2023 with input and feedback from Jan-Willem van Putten, Tim Bakker, Leon Lang, Teun van der Weij, Sam Bogerd, Daan Juijn, Stan van Wingerden, Jesse Hoogland, Alexandra Bos, Jelle Donders, Marieke de Visscher and James Herbert. Thank you. Any mistakes in this update are ours.