Technical governance

Why work on technical governance?

Export controls on advanced AI chips enjoy overwhelming bipartisan support in Washington. In January 2026, the US House voted 369–22 to close a loophole that let companies access controlled chips remotely via the cloud — a near-unanimous result from a Congress that’s often sharply divided.

But agreeing on the goal turns out to be the easy part. When the US first restricted chip exports in October 2022, the rules drew a line based on performance: above a certain threshold, restricted; below it, not. Within months, tech giant Nvidia had a new chip on the market for Chinese customers — built to the same underlying design, but with one specification altered just enough to fall under the line.

This is a common problem in AI governance. Even when policymakers agree on what they want — slow a rival’s access to compute, hold an AI company to its safety commitments, ensure a model has been properly tested before release — whether they get it depends on technical details: how you define a ‘high-performance’ chip, how you verify a safety claim, what a model evaluation can and can’t tell you. To make these policies actually work, we need people who understand the technology well enough to make a policy’s intent match its actual effect — and there currently aren’t enough of them.

Making advanced AI go well means solving a lot of hard technical problems. Some of these problems involve how the models themselves are built, but others are about governance and policy: figuring out what regulations we should have and how we can enforce them. Many of the most promising AI policy ideas, like compute governance, information security standards for frontier model weights, and model evaluations for dangerous capabilities, depend on technical work to figure out how effective they would be and how best to implement them.

Technical governance work spans the full public policy pipeline:

• Generating policy ideas that are grounded in how AI systems actually work
• Assessing technical feasibility — whether a regulation can actually be enforced, a safeguard can actually be tested, or a standard is well-defined
• Implementing policy once it exists, including building the evaluations, audits, and technical infrastructure that make policies work in the real world

There are too few people with technical skills in governance and policy work. Most policy organisations need more staff who can read a machine learning paper, understand a hardware supply chain, or assess a model evaluation report — and there are far more roles open than people qualified to fill them. If you have a technical background and you’re willing to learn the policy side, you may have a lot of good options to choose from.

Would you be a good fit?

There’s no single profile for technical governance — some roles are essentially research jobs, while others are about explaining technical ideas to non-technical decisionmakers, and others are closer to traditional policy or bureaucratic work. Several experts we talked to emphasised that, while some technical background is important, being a ‘bridge’ between technical and policy conversations is one of the most useful skills in this career.

What skills are needed to succeed?

For pretty much any technical governance role, you’ll want:

• Quantitative skills. Comfort with numbers, data, and technical reasoning is the baseline.
• An understanding of how AI works. You don’t necessarily need to be able to train a frontier model from scratch, but you should understand how modern AI systems are built and be able to follow ML research papers.
• Programming skills, especially in machine learning. This can come from a job, a university degree, an independent course, or self-study. That said, if you have substantial skills in other relevant areas, there are technical governance roles you can do with much less technical depth.
• The ability to ‘translate’ technical topics for non-technical people. Most decision makers are smart generalists, not specialists. They’re looking for people who can take a technical concept — say, why model weights are hard to secure, or what a compute threshold actually means in practice — and explain it clearly to a senator’s staffer, a policy director, or a reporter.

You’ll also need different skills depending on what type of work you want to do:

• Technical governance research requires many of the same skills as AI safety research.
• Policy work, whether in government or in less technical policy organisations, places equal importance on political and bureaucratic skills. The details depend on the specific role, and who your colleagues are — if the people you work with most closely are more technical, you may have a team culture where political skills matter less. But for many roles, you’ll need to be able to navigate institutions, build relationships, and work effectively with people who don’t share your technical background.

What experience is useful?

Most of the background that helps one become an AI safety researcher is also helpful here — but for many roles, you don’t need the same level of technical depth. Even people with technical credentials that they may regard as fairly modest — such as a computer science bachelor’s degree or a master’s in machine learning — often find their knowledge is highly valued in DC.

Useful kinds of experience include:

• A quantitative background is most important, especially in computer science or machine learning.
• Experience working in technical AI roles, or any policy role. Experience on both sides is ideal, but not required.
• An advanced degree in a technical field is helpful (or necessary) for some roles, but you can do many of these jobs without one.
• Publications and connections that establish you as an expert in a relevant technical policy area are helpful. This especially matters in DC and other government settings, where being known as a credible voice on a specific issue (compute, evaluations, infosec, hardware, etc.) can help open doors. A single piece of substantive, widely read analysis on a specific technical policy issue — compute governance, evaluations, infosecurity, hardware — can establish you as a credible voice quickly, even without an extensive CV.

What else do you need?

Some roles in the US government require US citizenship, and some — especially in military or intelligence settings — require a security clearance, which is hard to get and takes time. Non-US citizens can still pursue many technical governance paths, but the most security-sensitive ones will be off the table.

Downsides to working on technical governance

• Lower pay than many technical jobs. Frontier AI companies pay technical staff much more than policy organisations, think tanks, or governments do. The gap can be substantial, and it widens with seniority.
• Less job security than many technical jobs, especially for roles that depend on government funding, political appointments, or short-term contracts.
• Some governance roles are less directly ‘technical’ than they sound. Depending on the role, you might spend more time talking and writing about technical subjects than building things. Some people find this frustrating; others find it suits them better than full-time engineering.

Examples of people pursuing this path

Top organisations

The nature of technical governance work varies between employers — what you’d do at a third-party auditor looks quite different from what you’d do at a frontier AI company or in a government regulatory body. Below are some of the main categories:

Third-party auditors

Auditors check whether regulations are being followed. Governments often rely on third-party auditors in the private sector because it’s their best way to find people with the right expertise. This category will probably grow in the coming years as more AI regulations are implemented.

AI auditing work includes:

• Auditing model capabilities — evaluating whether a model has dangerous capabilities before it’s deployed; METR (formerly ARC Evals) has been a leader in this space
• Auditing compute use, infosec practices, and other operational requirements that AI companies may choose to audit now, or be required to by future regulation

Think tanks

Think tanks are independent research organisations that study policy problems and propose solutions. They sit outside government but are oriented towards influencing it — publishing reports, briefing officials, and translating complex technical issues into policy recommendations. Many of the field’s most important ideas about how to govern AI have come from think tank researchers.

• Center for Security and Emerging Technology (CSET) — produces policy analysis on AI and national security
• RAND — a broad public-policy research nonprofit whose AI work focuses on security, governance, and societal impacts
• Institute for AI Policy and Strategy — researches AI policy at the intersection of national security, compute governance, and international strategy
• GovAI — conducts AI governance research and fieldbuilding
• Institute for Law and AI — researches and advises on the legal challenges posed by advanced AI
• Centre for Long-Term Resilience — advises the UK government on extreme risks from AI, biosecurity, and risk management

Other nonprofits

Beyond think tanks, a range of nonprofits contribute to AI governance by other means — conducting empirical research on AI capabilities, building the field, advocating for safety-focused policy, and funding the broader ecosystem.

• Epoch AI — tracks and forecasts trends in AI development
• Apollo Research — conducts empirical research on AI model behaviour, focusing on deception, situational awareness, and dangerous capabilities
• Coefficient Giving — a philanthropic funder and advisor focused on high-impact giving across causes including AI risk (disclaimer: our main funder)
• Center for AI Safety (CAIS) — conducts technical AI safety research, fieldbuilding, and advocacy to reduce societal-scale AI risks
• Palisade Research — empirically studies dangerous AI capabilities (cyber, deception, shutdown resistance) and communicates findings to policymakers
• Civic AI Security Program — builds public understanding of AI capabilities and dangers through live demos and civil society briefings
• Secure AI Project — advocates for AI safety policy at the state and federal level (e.g. enforceable safety protocols for frontier developers)

Frontier AI companies

The top AI companies have internal teams doing governance-related work, including infosec, policy, model evaluations, and responsible scaling. This work is critically important: internal governance choices at frontier labs can shape what gets deployed, what gets shared with regulators, and how much information is available about model capabilities and risks.

That said, all the usual concerns about working at a frontier AI company apply. Read our career review on working at an AI lab before applying.

Governments

There are several ways to work on technical governance within governments:

• Technical regulatory bodies. In the US, key bodies include parts of the Department of Commerce, particularly NIST (which produces technical standards and benchmarks used to evaluate AI system safety and performance) and the Bureau of Industry and Security (which handles export controls, including the chip export controls that are central to current US-China AI policy). We expect to see many new hardware-focused regulatory roles in the next few years — see our career review on AI hardware expertise for more.
• Intelligence-related projects, often via Systems Engineering and Technical Assistance (SETA) contractors. You can’t do this work without a security clearance, which is hard to get — but if you already have one, or might be able to acquire one, intelligence work is itself a potentially valuable path, and having a security clearance will open up other career opportunities later on.
• Technical advising to help government decision makers make good calls on technical questions. The amount of technical expertise you need for this is lower than many people imagine.

Next steps

If you’re ready to apply for jobs

See our job board to search for roles and start applying.

If you need to build career capital

• A university STEM background is a good starting point. Computer science and machine learning are most directly relevant, but physics, math, electrical engineering, and other quantitative fields can all set you up well.
• Learn the basics. If you’re new to technical AI safety, our AI safety research career review explains how to get up to speed on the technical fundamentals. Most of that advice applies here too.
• Explore your fit at different points along the technical-policy spectrum. Try some highly technical work. Try a policy internship. Try a job in industry. The more things you sample, the faster you’ll figure out your fit for different options, as well as what you actually enjoy.
• Gain deeper expertise in a relevant topic. You might want to focus on hardware or information security — both are areas where deep technical expertise is unusually valuable for governance work. Getting experience in political systems and how the government actually works will help you in many technical governance roles.
• Apply to fellowships and internships. Policy fellowships and internships are among the best entry points into government and policy work, especially for people coming from a technical background.

Learn more

Top recommendations

• BlueDot Impact’s selective Frontier AI Governance course can be taken full time (six days in total) or part time over a longer period so that it fits alongside your job or studies. It covers the frontier AI landscape, key governance debates, and how to find your own path into the field, and is aimed at people seriously considering AI governance work. Alumni have gone on to work at places like RAND, CSET, IAPS, GovAI, CAISI, and frontier lab policy teams.

• Emerging Tech Policy is a free career resource website run by the Horizon Institute for Public Service. It covers pathways into policy, profiles of key institutions (Congress, think tanks, federal agencies), and guides to specific policy areas including AI, biosecurity, and cybersecurity. It’s aimed at people exploring or actively pursuing careers in tech policy, and is particularly useful for understanding how different institutions work, where you might fit, and what you might want to read next.

Further recommendations

• Career review: AI policy and strategy research
• Career review: AI policy in the US government
• Podcast: Dean Ball on how AI is a huge deal — but we shouldn’t regulate it yet
• Podcast: Allan Dafoe on why technology is unstoppable & how to shape AI development anyway
• Podcast: Daniel Kokotajlo on what a hyperspeed robot economy might look like
• Podcast: Lennart Heim on the compute governance era and what has to come after and Lennart’s technical AI governance blog
• Podcast: Sella Nevo on who’s trying to steal frontier AI models, and what could they do with them

Acknowledgements

We thank Lennart Heim and Aris Richardson for comments on drafts of this article.

Read next:  Learn about other high-impact careers

Want to consider more paths? See our list of the highest-impact career paths according to our research.

Continue →