AI safety in national security
Review status
Based on a medium-depth investigation
Table of Contents
On March 23, 2003, a Patriot missile battery in northern Kuwait identified an incoming anti-radiation missile. The crew had just a few seconds to decide whether the threat was real. They fired before the ‘missile’ could get any closer — and killed both people aboard what turned out to be a British aircraft heading back towards friendly airspace.
The missile battery did have humans in the loop, but the crew was working with a system that was right almost all the time, under time pressure that made it hard to assess the specifics. The Army’s own review found that crews had been trained to trust the system and given almost no basis on which to doubt it.
That was 2003 software, and the AI systems of today only make the problem harder. Today’s systems can do far more than any technology in 2003 could, so the temptation to let them function more and more autonomously will continue to grow. As of 2026, frontier AI companies are working with the US military on integrating AI tools. Many experts are worried about what might happen if that integration isn’t done safely.
As the AI tools used by militaries become both more powerful and more autonomous, the scale of potential failure becomes increasingly large. Militaries already use autonomous fighter jets and drones. A growing share of intelligence analysis is based on AI, and cyberwarfare is speeding up as AIs become more competent hackers.
Most people who want to work on risks from advanced AI don’t consider careers in national security. For some, this is a deliberate decision, because they don’t want to work in an industry that inherently includes violence and could cause significant harm to innocent people, or because they don’t support integrating AI into the military and don’t want to be involved in any way. We think both of these can be good reasons to choose other paths.
But we also think that some of the most dangerous uses of AI will arise in the context of national security, and that having people in security roles who are knowledgeable about risks from advanced AI could avert some of the worst possible AI outcomes. And because many people see national security work as a form of public service, and a chance to make their country and the world safer, we want to make it clear that we see a strong path to impact in this field. These roles aren’t for everyone, but if you’re a good fit, they are worth considering.
Note: this article is focused on national security roles within the United States. If you’re in a different country, some of the same logic applies, but we haven’t researched those roles.
Why work on national security?
The AI safety case for national security work comes down to three things: the stakes of how AI is used, the shortage of safety-minded people working on reducing risks in this area, and the leverage that comes with the US government’s vast resources and the power it controls.
The work can also be personally fulfilling. Many people who work in national security chose it because they want to serve and defend their country. They often see the work as morally complex, but still good overall. This article is focused on AI risk, but patriotism is an additional reason for some people to consider this path.
National security includes some of the most dangerous AI deployments
National security applications are among the most consequential — and potentially most dangerous — ways AI is being deployed. These include (but aren’t limited to) autonomous and semi-autonomous weapons, offensive and defensive cyber capabilities, intelligence analysis that feeds decisions about the use of force, and surveillance tools that could shift the balance of power between governments and their own citizens.
We’ve been living in a period with no major wars between global powers, and nuclear weapons haven’t been used since 1945. But we almost lost this relative peace a few times because of automated information processing systems. For instance, in 1983, when Lieutenant Colonel Stanislav Petrov saw that early detection systems were reporting five incoming US missiles, he was supposed to pass that information up the chain of command, which could have led to a Soviet strike on the United States in retaliation. Petrov took the time to think about it and reasoned that a real US strike would likely have involved hundreds of missiles, not five, and reported a system malfunction instead.
As AI systems become more integrated into our national security apparatus, a modern-day Petrov would have a harder time making that call. Of course, today’s systems are smarter than those from the 1980s and less likely to make simple mistakes, but in some ways, that makes the problem even harder: if a very competent AI system says that it’s important to launch a counterstrike immediately, it may be hard for a human to question that recommendation and notice whether the AI made a mistake or is compromised or misaligned.
The sheer volume of AI-generated information may also make decision making more difficult than it was for Petrov: in September 2026, CNN reported that a source claimed an intelligence report generated using AI “almost started a war” when it misidentified a Chinese ship’s cargo as containing components of a nuclear weapon. In this case, the mistake was identified before any serious harm was done, but as humans deal with increasing amounts of AI analysis, they’ll have a harder time digging into details to correct mistakes.
If AI systems are implemented within command and control, humans will likely end up with fewer opportunities to notice and stop an erroneous process before it leads to a catastrophic outcome. The systems we’ve developed have held up so far, and none of our mistakes have led to a global nuclear war. But if AI introduces new types of mistakes and new possible responses, this balance may become much less stable. There’s already more information available than humans can look at, so analysts will rely on automated processing to help them make sense of the data, making conclusions hard to verify, even if AIs aren’t making decisions independently.
Other AI capabilities could also become extremely dangerous in a national security context. AI systems are quickly developing superhuman hacking abilities, and hacks against critical systems could become increasingly common as powerful cyberwarfare tools become accessible to groups without much money or technical expertise. The same is true for biological weapons and other extremely dangerous technologies: it may become possible for relatively small-scale actors to launch attacks that only major powers could execute today.
To keep people safe as these technologies develop, we need people within the US national security apparatus who understand AI and are concerned about its possible risks. The US government has the resources and power to defend against these risks when they come from other actors, so it’s one of the best places in the world to work on stopping AI misuse and defending against some types of rogue AI behaviour. At the same time, the US security establishment is also using advanced AI models in new and potentially dangerous ways, so it’s important for the people within it to be knowledgeable about the risks and to work to make sure AI is used safely.
National security needs more people who care about AI safety
The kind of person who worries the most about catastrophic risks from AI is rarely the kind of person who pictures themselves at the Department of War. People motivated by AI safety often cluster in research labs, academia, and a handful of policy organisations, which means the military, intelligence, and other agencies making some of the highest-stakes AI decisions can’t easily hire people with strong AI fluency and commitments to making AI go well. Although there are people focused on AI safety within many of these organisations, and there will likely be more as the issue becomes more mainstream, we still need many more talented people who care about AI safety working on national security.
Government money shapes what AI companies build
Defense and national security account for an enormous share of US federal spending — roughly half of the discretionary budget. And because the government spends so much money on defense, it has a huge amount of influence on corporate R&D. Procurement requirements, testing standards, and contract terms shape what AI companies build, what they’re allowed to deploy, and what safety practices become normal — often more directly than regulation does. The government’s role as AI’s largest and most demanding customer gives the people writing those requirements leverage over how the technology develops, including at the frontier AI companies now seeking defense contracts.
In early 2026, when Anthropic and the US Department of War had a public conflict over the licensing terms for military uses of Anthropic’s AI model, Claude, some of their central points of contention were reportedly about using Claude for fully autonomous weapons and for mass domestic surveillance. Anthropic wanted to bar the use of Claude for these activities; the Pentagon wanted to be able to use it for any lawful purpose. When Anthropic refused, the Pentagon designated it a “supply chain risk,” a label usually reserved for companies tied to foreign adversaries, and later came to an agreement with Anthropic’s competitor OpenAI. Few customers could have that kind of leverage over a company. That means whoever decides what the government asks for, including which safety features it requires or rejects, could have substantial influence over how this technology develops.
Would you be a good fit?
There’s no single profile for careers in national security — the field includes intelligence analysts, programme managers, think tank researchers, and technical evaluators, and those jobs reward different strengths. But they share more than the job titles suggest: nearly all of them involve working inside large, hierarchical institutions, communicating across a technical / nontechnical divide, and earning trust in a culture that is often sceptical of new ideas and changes to established procedure.
What skills are needed to succeed?
Most AI-relevant national security roles call for some combination of:
- Social and political skills: National security institutions run on trust and informal networks even more than the rest of the government. Who you work with tends to matter more than your job title, so a relatively junior person whose mentor has an influential position may be able to achieve a lot. Compared to the tech industry, DC tends to be relatively traditional, so conventional professional norms (dress, behaviour) are important for many roles. But context matters a lot: roles that are more technical and are farther from elected officials may have more room for personal expression.
- Technical skills: Some amount of technical skill is almost always helpful, but the degree of helpfulness varies — and you may need much less skill than you’d expect. If you’re used to the hiring standards at leading tech companies, it’s important to remember that your skills may go much further in organisations with few staff in technical roles. You don’t need to be able to build AI systems (with rare exceptions), but you’ll usually need to be comfortable reading and understanding a paper and explaining it to someone less technical than you are.
- ‘Translating’ technical concepts for people with varied backgrounds: In many roles, some of your most important work will involve helping technical and nontechnical people communicate, such as by helping an engineer understand what a policymaker needs and helping a policymaker understand how a system works. This means knowing enough to understand both sides of the conversation, and conveying information in simple language without slipping into condescension.
- Briefing: This means being able to compress a detailed report or data set into a single page, or a few minutes of discussion. Decisions in this world are made by busy people, and the person who can summarise well shapes what gets understood.
- Writing: Policy writing is a different genre from academic writing: bottom line up front, short, and with clear action items. If you prefer to write long essays full of caveats and counterarguments, you might not enjoy writing in this style.
What experience is useful?
People enter this work from many directions, but these are particularly useful:
- Military service: A service record is useful for credibility and institutional knowledge, and many federal roles give preference to veterans. However, service isn’t required; most people in these roles don’t have it. Because AI progress is happening so fast and attaining high military rank can take many years, we don’t recommend that people enter the military with the goal of affecting AI policy, but if you already have military experience, it can certainly help.
- Graduate degrees: Policy master’s degrees are particularly useful in DC, as are law degrees and PhDs in relevant fields.
- Intelligence or other government experience: Time in any policy role builds networks and an understanding of how institutions work, and if you get a security clearance in the process, that opens up a much wider set of roles.
- Critical languages: Knowing a language that’s relevant to key international AI safety issues is necessary for some roles and useful for many others. Mandarin and Russian are particularly helpful for some analysis and intelligence roles, although there are also many roles in these areas that don’t have any language requirements. For most people, it’s probably not worth spending a lot of time learning a language in order to work on AI safety within national security. But if you do have this background, it may help you.
- Legal fluency: You don’t need a law degree, but it’s often helpful to feel comfortable reading statutes, regulations, and procedural documents. The culture tends to be conservative, and people often assume that statute or policy requires something to be done as it always has been. If you’re comfortable reading legal jargon and drawing your own conclusions, you may be able to help people find other options that they didn’t know would be allowed.
What else do you need?
Beyond skills and experience, many national security roles come with practical requirements that are important to consider as you decide whether these roles could be a good fit for you:
- For most government roles, you’ll need US citizenship. Most roles at think tanks and research organisations can be done by foreign nationals.
- Many roles require a security clearance, and the process can be complicated and invasive. On the other hand, an active clearance compounds as a career asset: it makes you dramatically easier to hire for the most consequential roles. If this path appeals to you at all, it’s worth positioning yourself to start the process early — but keep in mind that clearances expire if they go unused for a few years. (More on clearances from Emerging Tech Policy Careers.)
- Some roles in the executive branch and most roles working for members of Congress effectively require party affiliation, and prior associations with one party (including donations) can lock you out of some roles under the other.
- Most of these jobs require you to live in or near DC, or a regional hub for your agency. Remote work is limited, and often impossible for roles that require a security clearance.
- Finally, you’ll need to be comfortable with the basic premise of the work. You don’t have to be enthusiastic about defense institutions, or think that the American military does only good things, but you do need to be comfortable working alongside the military. In addition, these are jobs serving the United States. If you see reducing AI risk as part of serving your country, this path could be a great fit. On the other hand, if it’s important to you to work in an environment that prioritises all of humanity equally, regardless of where they live, this path might not be the right fit for you, even if the impact case seems appealing.
Downsides to working on national security
Serious risk of doing harm
National security institutions are powerful, and their mistakes are hard to reverse. Working inside them could position you to prevent serious harm — or to contribute to it. You might do harm by:
- Helping to build or legitimise dangerous capabilities that wouldn’t otherwise have been fielded, or fielded as quickly
- Directly making a choice that contributes to a dangerous accident
- Making harmful uses of AI more likely by allowing institutions to ‘safety-wash’ their dangerous AI deployments by pointing to supervision from safety experts like you
- Entrenching AI systems in a way that makes it harder to stop using them later on, if they become dangerous in the future
The fact that so much of your work may be classified makes this problem worse. If you’re someone who usually talks difficult decisions through with friends or mentors, the level of required secrecy may make it hard to get enough perspective to notice if the work you’re doing is no longer addressing the problems you want to solve.
Personal risks to you
Most of the roles discussed in this article involve sitting at a desk in the United States and never facing significant physical danger. But if you’re embarking on a career in national security, you should take into account that some of these roles (including technical roles in field intelligence) put people at risk of being injured or killed in the line of duty. In addition, some roles may expose you to personal harms like hacking attempts or other cyberattacks.
Beyond those dangers, working in national security exposes you to moral and psychological risks that are different from those in most other areas. For example, you may work closely with field operatives, who may be injured or killed as part of their duties. You may also face moral injury if you participate in projects that harm or kill other people, especially people who are not themselves dangerous. The secrecy required by many of these roles can make these risks harder to cope with, because you may not be able to process what you’re facing with the people in your life the way you would in most jobs.
Risk of spending time on work that doesn’t matter
More prosaically, a lot of work in the national security establishment may simply not be very useful. These are large bureaucracies, and you won’t fully control what you work on. Projects can lose focus and become much less impactful than they initially seemed. It’s also common to spend a long time on a project that ends up getting cancelled, reorganised, or defunded, or to invest time and effort into building professional relationships that become irrelevant when a new administration comes in with new personnel and priorities.
Long and opaque hiring process
The federal hiring process can often be slow and hard to predict. Because of civil service requirements, the hiring manager for a role is not always the person who actually decides who to hire, so it may be harder to straightforwardly use your resume and skills to get a job in the federal government than in most industries.
In addition, many national security–relevant roles require a security clearance. Once you have one, you can often move into subsequent roles without repeating the whole process, so having a clearance is a useful asset for future roles; you can start working sooner if you have a clearance from a prior role. Because of this, some roles will heavily favour people who already hold an active security clearance, which can make getting into your first national security role harder.
For roles where you need to get a security clearance, the timeline isn’t up to you or the people you work with. For a Secret-level clearance, which is the most common level, an average case will take about 4–5 months, and a more complicated one can take 6–12 months. Top Secret clearances typically take about eight months, and cases with complicating factors could take more than a year.
If you fail a clearance application, you’ll have a harder time getting a clearance later. Because you can generally only be sponsored once a year, it’s better to avoid applying for a clearance unless you feel very confident you’ll pass. If your application looks likely to fail, you can sometimes withdraw it to avoid having a denial on your record, but it’s best not to rely on that option.
The form used in the clearance process asks, among other things, whether you have used any federally illegal drugs (including marijuana) in the past seven years. Answering “yes” will not necessarily disqualify you, but it can, and ongoing regular drug use likely will. It’s important to answer truthfully and fully. In most cases, not having used drugs in the past year is a mitigating factor, so don’t automatically screen yourself out of these roles if you have used drugs in the past. However, these roles probably won’t be a good fit if you regularly use marijuana or other federally illegal drugs and intend to continue doing so.
Because getting a clearance is such a long and involved process, several experts we talked to see it as a significant barrier to working on national security. It makes it harder to break into national security work for the first time, and it means that, if you do get a role, there will likely be at least four months before you can start working with classified information (although some roles will let you start earlier with limited information access). If your timelines are short or you’re eager to start working on important things soon, the clearance process may be a significant downside of these roles compared to other policy work.
Lack of AI safety community
We think some of these roles may be especially valuable because there aren’t many AI safety professionals doing this work. But the flip side is that you might be the only person in your department who is knowledgeable about the full range of risks from advanced AI; your colleagues might have negative opinions about the AI safety community, or not believe that the risks are actually plausible. You might find this lonely and demotivating, or find yourself drifting towards the opinions of the people around you and thinking less about AI safety over time.
Lifestyle and expression constraints
Some of these roles — especially in the intelligence community — heavily restrict your actions and your freedom of expression. You might spend most or all of your workdays in a SCIF without your phone; anything you want to publish online will probably need review; you’ll have few chances, if any, to work remotely; you might even be asked to take regular polygraph tests.1 You might not even be able to tell important people in your life about your work. For more on this, see Emerging Tech Policy Careers’ page on the downsides of intelligence work, some of which also apply to other national security roles.
Top organisations
The institutional landscape of national security shifts constantly — offices are reorganised, programmes are renamed, and hiring rounds open and close unexpectedly when a new administration or budget takes hold. We’ve used this section to map some important agencies, but check their websites for updates if you plan to apply.
Note that this list doesn’t include roles limited to military officers or enlisted personnel. If your primary goal is to improve AI policy or safety, we think you’ll have better odds working outside the military, since most policy is made by civilians. However, if you’re already in the military, especially if you’re relatively high in the chain of command, that experience will likely be valuable to you in many national security roles, and laws giving preference to veterans may also help you get roles that might otherwise be inaccessible.
Government
- Department of War: The largest AI adopter in government by far, spanning everything from the Chief Digital and Artificial Intelligence Office to individual service programmes. Requirements, testing, and acquisition decisions made here shape what gets built and used.
- Department of Commerce: Home to two very different offices that both matter enormously.
- Bureau of Industry and Security: Writes and enforces the export controls on AI chips — arguably the US government’s most consequential lever over frontier AI development, run by a strikingly small team.
- National Institute of Standards and Technology’s Center for Advancing Innovation and Standards for Super Intelligence: Develops AI standards and evaluations and is one of the few places in the government doing technical work on frontier model capabilities.
- Executive Office of the President: Coordinates AI and national security policy at the highest level, particularly through the National Security Council and Office of Science and Technology Policy. Realistically, you won’t get hired here early in your career; you’re more likely to be detailed or appointed to the Executive Office after building a track record elsewhere. These roles are likely to be very impactful if you can get them.
- The intelligence community: This includes the CIA and NSA, as well as agencies that are less well known to the public, including the State Department’s Bureau of Intelligence and Research, the Defense Intelligence Agency, and the Office of the Director of National Intelligence, plus organisations like the Intelligence Advanced Research Projects Activity that fund frontier-relevant research. Many of these are analysis roles where AI fluency is scarce and valued, but they can have some of the highest lifestyle costs (see downsides).
- The enforcement side of government: Often less discussed, but this is where controls become real. Agencies include the State Department’s Directorate of Defense Trade Controls (arms export licensing), the Treasury’s Office of Foreign Assets Control (sanctions), the Energy Department’s National Nuclear Security Administration (nuclear security, with growing AI work), and the Department of Justice and FBI (investigating violations). Rules that nobody enforces don’t constrain anyone, and these offices are chronically short of people with expertise in frontier AI.
Think tanks
- Center for Security and Emerging Technology: A Georgetown University research centre that has become one of the most widely cited sources of data-driven analysis on AI and national security, including compute, the AI workforce, and US–China competition. Its work is read closely inside government, and several of its researchers have gone on to senior policy roles.
- RAND Corporation: One of the oldest defense research organisations in the US. It runs several federally funded research and development centres (FFRDCs) for the military, so its research often goes directly to the people making decisions. Its Center on AI, Security, and Technology focuses on national security risks from frontier AI.
- Center for Strategic and International Studies: A large DC think tank covering defense and foreign policy. Its Wadhwani AI Center works on AI policy, export controls, and US–China tech competition, and its reports and events are a common way for new ideas to reach policymakers.
- Center for a New American Security: A defense-focused think tank with close ties to the national security establishment, and a steady flow of staff between it and senior Pentagon and White House roles. Its technology and national security programme covers military AI, AI safety and stability, and compute.
- Special Competitive Studies Project: Founded by former Google CEO Eric Schmidt to carry on the work of the National Security Commission on AI. It focuses on US technological competitiveness, especially against China. It puts more weight on winning that competition than on reducing risk, so its emphasis is different from most of the other organisations listed here.
- Institute for AI Policy and Strategy: A smaller, newer think tank focused on reducing risks from advanced AI. Its research covers compute and chip export controls, US–China dynamics, and the security of frontier AI systems.
Research organisations
- Centre for the Governance of AI: Conducts AI governance research, including research into threat modelling, military and geopolitical implications of AI, and AI adoption in the military (as well as other research areas that are not directly about national security).
- Federation of American Scientists: Science policy nonprofit that works on AI, emerging tech, and nuclear risk, and helps place technical experts in government.
- Institute for Progress: Think tank focused on innovation policy, including research on AI security.
FFRDCs and university-affiliated research centres (UARCs)
These centres do long-term research and analysis for government agencies. People who work at these organisations are often issued security clearances and therefore can have closer access to government decision making than independent think tanks. Note that all of these organisations do a wide variety of research, some of which is not relevant to AI safety and some of which may be actively harmful, so it’s important to find out the details of the individual project you are applying to work on — we would not recommend all jobs at these organisations. A few examples:
- RAND: Operates several FFRDCs for the Department of War and Department of Homeland Security, alongside its independent research.
- MITRE: Nonprofit that runs FFRDCs across defense, cybersecurity, and civilian agencies, with a growing body of AI assurance and security work.
- Department of Energy national labs (e.g. Lawrence Livermore, Los Alamos, Sandia, and Oak Ridge): Home to some of the government’s largest computing resources; increasingly involved in AI for science and in evaluating frontier AI for national security risks.
- Johns Hopkins Applied Physics Laboratory: A UARC doing applied defense research, including on AI.
Key roles to aim for
Because the national security landscape is so vast and changes so quickly, we aren’t listing specific roles. Instead, here are some tips that might help you decide how impactful a given job might be:
- Does the role help you build career capital? Clearances, interagency experience, and a reputation for reliable technical judgement all compound. Especially early in your career, look for roles that leave you with at least one of these even if the specific programme fizzles.
- Who will you be working with? Who you work with matters more than your job title. A relatively junior role near people making consequential decisions is often better than a senior-sounding role without the same connections, because you may have the opportunity to move with your boss if they get promoted.
- Does it touch the interagency process? All else equal, roles where you engage with how decisions actually get made across agencies — drafting policy documents that circulate or staffing interagency working groups — tend to build more leverage and career capital than roles confined to one office’s internal work. However, there are many individual roles that are valuable despite a lack of engagement with interagency processes.
- Does this agency already have people who moved into national security or policy work mid-career? If it has none or very few, that may be a sign that it’s less open to hiring those people, which may make it both a harder place to get hired and a harder place to have an impact on the most important AI problems.
How to enter if you’re early career
Early in your career, your main goals are getting a foot in the door and building the credentials, such as a clearance and institutional experience, that open up more senior roles later. A few ways to get started:
- Apply for fellowships. The Horizon Fellowship, TechCongress, and the AAAS Science and Technology Policy Fellowships all place people into government and policy roles, and they come with cohorts and networks that help you build connections and institutional knowledge.
- Start the clearance process early. If this path appeals to you, look for a role where you can start the clearance process.
- Start applying! Many agencies have entry-level positions, and think tanks often hire research assistants and other people in junior roles.
How to enter if you’re mid-career
Switching into government and policy roles mid-career can be difficult, because you may be too senior for most entry-level jobs to be a good use of your time, yet not have the policy experience to get a more senior role. National security roles in particular may rely on tacit knowledge and networks that many people build up over the course of their careers. These steps can help you make the switch:
- Apply for fellowships. Policy fellowships can help people make this transition:
- The Horizon Fellowship trains and places fully funded fellows at host organisations in emerging tech policy, including in national security positions.
- If you’ve started a career in academia, the Belfer Technology and Geopolitics Fellowship for pre- and postdoctoral candidates may be a good fit.
- The US Tech Force is a two-year programme for engineers to work in government, including as part of the War Force.
- Build your network and practise pitching yourself. One of your challenges will be building a professional network that helps you get roles. It helps to have a clear answer to “What do you do?” that explains why your skills and background would be valuable outside the context you’ve worked in so far.
- Consider a nonpolicy role as a way in. In some circumstances, taking a nonpolicy role can help you break in, but be wary of roles that don’t have a clear path towards the work you actually want to do.
- Use resources for a career switch. If you’re coming from tech, the site Tech to Gov and Public Interest Technology Resources both have helpful guidance and recommendations.
- Consider a graduate degree. Graduate degrees can be particularly helpful for landing jobs in the federal government, both because they may provide knowledge and connections that will help in your work, and because federal hiring rules give preference to people with graduate degrees in some situations. If you’re thinking about getting a graduate degree for this purpose, a policy master’s degree may be worth considering.
Next steps
If you need to build career capital
- Begin building your network. Many roles in this field are filled through networks and word of mouth. See our networking advice. It can also help to spend some time in DC if you can, and go to professional events to introduce yourself and talk to people working in areas that might interest you.
Publish writing on relevant topics (including by starting a blog). Writing some blog posts or publishing an article is a surprisingly good way to get known as someone with interesting and well-informed views on a topic.
Build technical literacy and practise explaining things. You don’t need to be a technical AI safety researcher, but for many roles, you do need to credibly demonstrate that you understand AI systems (technical degrees, work at AI companies, evaluations experience, and serious coursework all help). Many people coming from more technical communities find that the amount of technical background that ‘counts’ in DC is much less than they expect.
If you want to learn more
Top recommendations
- Emerging Tech Policy Careers: A free career resource website run by the Horizon Institute for Public Service. It covers pathways into policy, profiles of key institutions (including the Department of War and the intelligence community), 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.
- How AI could transform the nature of war: Paul Scharre, a former Army Ranger who led the Pentagon team that wrote the US military’s first policy on autonomous weapons, discusses how AI could change warfare and what happens when military decisions start moving faster than humans can follow.
Further recommendations
- Career review: AI policy in the US government
- Career review: Technical governance
- Career review: AI security
- Podcast: Helen Toner on the geopolitics of AI in China and the Middle East
- Podcast: Sam Winter-Levy and Nikita Lalwani on how AI won’t end nuclear deterrence (probably)
- Podcast: Jake Sullivan on Navigating AI Uncertainty and Managing Competition with China (on The Cognitive Revolution)
Speak with us
If you think this path might be a great option for you, but you need help deciding or thinking about what to do next, our team might be able to help.
We can help you compare options, make connections, and possibly even help you find jobs or funding opportunities.
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.
Acknowledgements
We thank Chris Byrd, Dave Kasten, and Trinity Dial for advice and comments on drafts of this piece.
Notes and references
- Many people find the polygraph process (often part of the hiring pipeline for intelligence roles) particularly stressful, even if they haven’t done anything wrong and are telling the truth: people with nothing to hide sometimes even fail a polygraph and need a retest, possibly because stress signals can mimic lying.↩