AI Safety Career Profiles
Who built this

Hey! I'm Manon. Over the past year I've been starting and running AI Safety Saarland, organising the Safe AI Germany Incubator, mentoring local AI safety groups around the world, and TAing at an ML4Good bootcamp. This summer I'll be a resident at the Generator.

When I started out in AI safety myself, the only career paths I really saw were technical research or policy. As I got more involved with field-building, I realised there are many more roles and that people filling them are in great demand right now. But it wasn't clear to me what those roles actually entailed. I ended up talking to grantmakers and research managers at EAG SF and got a much richer picture for myself.

When I taught at ML4Good this May, I saw how many people there also couldn't see career options beyond technical or policy research. That's what motivated me to build this.

LinkedIn manonkempermann.eu manon@kempermann.email
The tool

This is an independent community resource, not affiliated with any organisation. It aims to give anyone exploring AI safety a concrete, interactive way to discover roles they might not have considered, based on how they actually work and think, not just their technical background.

The tool works by matching users' working preferences against role profiles built from professionals' survey data, the responses collected here. The more professionals contribute, the more accurate the profiles become.

Users answer around 30 questions about how they work. Their profile is then matched against averaged professionals' role profiles to surface fit scores for each role, including ones they would never have thought to consider. Each role card shows a day-to-day description, surprising skills, who struggles there, and links to further resources.

The tool will be freely accessible to anyone, anywhere, and I'm actively working towards making it available through organisations that help people enter the field.

Most AI safety career paths are still invisible.

Many people entering the field still only see technical research or policy and governance work as the career paths in AI Safety. But the field needs and offers many more: research managers, program directors, educators, community builders, communications leads, operations generalists, chiefs of staff. While in recent months fellowships like the Generator Residency have popped up, these roles are still quite invisible to many newcomers. It might be easy to imagine what a technical researcher is doing day-to-day, but quite unclear what this looks like for grantmakers or research managers.

This tool aims to help overcome this information gap. It will give anyone exploring AI safety a concrete, interactive way to discover roles they might not have considered based on how they actually work and think, not just their technical background. Someone might go in assuming they want to become a researcher, and discover that research management or grant-making is actually a much better fit.

To build accurate role profiles, we need data from people working in those various roles like you. Your responses will be anonymised, averaged with others in similar roles, and used to populate the tool's role cards. The tool will be openly available as a community resource. This form takes about 12–15 minutes and your responses will directly help people aiming to contribute to the field understand what the day-to-day work in your role actually looks like and where they may fit best.

How is this different from 80,000 Hours career profiles?

80k's career profiles are excellent and I highly recommend them. But they are necessarily broad, describing paths at a high level, aimed at helping people choose between major categories of work.

This tool tries to go deeper in a different direction: the day-to-day work style of specific roles, the surprising skills they require, and who would really thrive versus struggle in them. It will interactively guide users through questions on their working preferences and match their scores to those of different career profiles. The goal is that people discover more types of AI Safety jobs they might be a good fit for.

Wouldn't this replace the good conversations people have at EAG?

No. EAGs are great places for people entering the field to learn about new career options, but this tool should actually help them make even more out of their time there.

If people don't know about roles or haven't even considered a second whether this could be interesting to them, they likely won't think to seek out a person with that role at EAG. This tool surfaces the full landscape early, so people arrive at those conversations with better questions.

Access to EAG and EAGx is also uneven. When organising my local university group, I often had people facing barriers based on nationality, background, geography, and timing. This tool will give people at any place and time a way to learn more about how they could contribute to AI Safety best.

How will this data be used?

All responses are anonymised and aggregated by role type. Nothing will be attributed to any individual or organisation. Role profiles in the tool will reflect averages and patterns across multiple professionals, not any single person's answers.

I am actively seeking collaboration with organisations that target people entering the field like introductory courses, fellowships, bootcamps, university groups, and other entry points to offer this as a shared community resource they can point people to. The tool itself will be free and openly accessible.

No data will be sold, shared with third parties, or used for anything beyond building and improving this tool.

Which role type fits you best?

Pick the description that is closest to your current role. If you wear several hats, choose the one that takes up most of your time. This helps us average responses across people doing similar work.

Research & technical

Strategy & funding

Policy & governance

Communications

Talent & education

Community & field-building

Org management & operations

None of the above

How your responses build the tool

This survey has two parts. Here is what each one is for, and how it shows up in the tool that people will use.

1

Profile questions · ~10 min · required

30 scale questions about how you actually work: your autonomy, collaboration style, output type, technical depth, and so on. Your answers produce a role profile: a set of scores across 7 dimensions that describes what working in your role is like.

We average these across everyone who identifies with your role type. The result is an holistic profile for that role. It is not a job description but a more representative picture of what the work actually feels like day to day.

When a user takes the tool, they answer the same questions about their own preferences. Their profile is then compared to all the role profiles, and the closest matches surface at the top, including roles they may never have considered.

Tool output: role matches (example)
87% Research manager Management
74% Program / fellowship manager Talent
61% Empirical safety researcher Research
2

Role description · ~5 min · optional

A short set of open questions about what your role is actually like: the day-to-day activities, the skills it requires, who would struggle in it, and common misconceptions. These feed the role cards users read when a role surfaces as a match.

This is the part that makes the tool useful and helps users look deeper into their best match. A user who has never heard of "chief of staff at an AI safety org" can read your words and understand concretely what that means and whether it appeals to them.

Tool output: role card (example)
Research manager
87% match

Day to day: reviewing research proposals, running team standups, unblocking researchers, writing internal strategy memos.

Surprising skill: knowing when not to weigh in, and learning to hold space for others to solve problems.

Who struggles here: people who need to be the one generating the ideas, or who find other people's problems less interesting than their own.

The profile questions (Part 1) describe your role as it is, not how you'd ideally want it to be. The more honest the input, the more accurate the role profiles and the better the tool works for everyone.

Work & intellectual style

Place yourself on each scale based on how your work actually is and not how you wish it were or how it is described in a job posting. 1 = fully left, 5 = fully right.

Please answer all highlighted questions before continuing.

Work Style

A1
In my work, my time is mostly self-directed. I set my own priorities with minimal fixed external commitments.
My time is largely structured by meetings, deadlines, and coordination with others.
A2
Most of my best output comes from long, uninterrupted solo blocks focused on one thing.
Most of my best output comes through frequent exchanges (conversations, collaboration, real-time problem-solving).
A3
I work mostly with a small, tight group within my immediate area.
I work across many different people, teams, and organisations outside my core area.
A4
I define my own direction: I decide what I work on and why.
I work towards goals that others set.
A5
The point of my work is my individual contribution. I am the one doing the work.
The point of my work is making others' contributions better. I multiply the impact of many people.
A6
I mostly produce one main type of output on a regular rhythm.
My tasks, stakeholders, and outputs shift substantially week to week.

Intellectual and research style

B1
I spend most of my intellectual energy generating novel frameworks or ideas.
I spend most of it synthesising, translating, or communicating ideas that others generate.
B2
I go very deep on one question or research agenda over extended periods.
I work across many questions or projects simultaneously, making connections between them.
B3
My core output is new knowledge (findings, papers, insights that advance understanding).
My core output is running systems (programs, tools, organisations, communities).
B4
Writing is a central output of my work (papers, memos, essays, reports).
Execution is the central output. The result is something that runs or keeps running, not something that's read.
B5
I need to understand something deeply before I can act on it with confidence.
I experiment a lot, observe what happens, and iterate from there.
B6
I work on things where feedback on whether they mattered comes on very long timescales (years).
I get relatively quick feedback on whether what I'm doing is working (weeks to months).

Relationships & communication

1 = fully left, 5 = fully right. 3 = neither applies, or it varies a lot.

Please answer all highlighted questions before continuing.

Relational and amplifier disposition

C1
Explaining things to others is peripheral to my core work.
Teaching, mentoring, and explaining is one of the most central and satisfying parts of my work.
C2
Managing others and developing their careers is a small part of what I do.
Managing people and actively developing their careers is a major and valued part of my role.
C3
Organising events, workshops, or gatherings is a minor or absent part of my work.
Organising and hosting events, workshops, or gatherings is a major part of what I do and something I get energy from.
C4
I maintain a small number of deep relationships within my immediate team or field.
I maintain broad relationships across many institutions and contexts. Being a connector is part of the work.
C5
When thinking about careers, I'm focused on my own trajectory.
Supporting other people's career development and navigation is a major part of what I do.
C6
I focus on a well-scoped project and deliver it.
I hold the broader picture and coordinate others toward it.
C7
Identifying, approaching, and hiring people into specific roles is not part of my work.
Identifying, approaching, and hiring people into specific roles is a significant part of my work.

Communication style

D1
When explaining something complex, my default is narrative and storytelling.
My default is structured argument and formal prose.
D2
My outputs are aimed at broad general audiences who didn't actively seek them out.
My outputs are aimed at specific decision-makers or specialists who need them to act.
D3
Building a public presence is part of my role.
I work mostly behind the scenes; attribution matters little and I prefer it that way.
D4
Most of my communication is internal (within my organisation or close network).
Much of my communication is outward-facing (external relationships, public representation).
D5
My outputs stand on their own. People read or use them without me having to actively sell them.
Actively persuading, pitching, or mobilising specific people is central to how I have impact.

Strategy, judgment & technical depth

1 = fully left, 5 = fully right. 3 = neither applies, or it varies a lot.

Please answer all highlighted questions before continuing.

Strategy and judgment

E1
I work on a well-defined problem within an established direction.
I spend significant time deciding which problems or directions deserve attention and resources.
E2
I focus on my own output without thinking much about what the field as a whole needs.
I regularly think about where the field is going and what it needs next.
E3
How organisations are structured and how decisions get made is not a significant part of my thinking.
Organisational design and how decisions get made is something I think about and actively work on.
E4
I work on things with fairly well-established value.
I regularly take high-risk, high-leverage bets with genuinely uncertain outcomes.

Technical depth

F1
Strong ML or formal methods is a core skill central to my daily work.
I keep up-to-date with the technological development, but ML and formal methods are not my primary tools.
F2
Writing good, robust code and building well-engineered systems is a core craft I care about.
Writing code is not part of my work.
F3
If I engage with technical papers, I do so mostly for high-level findings and implications.
I engage with technical papers for their content (methodology, proofs, and implementation details).
F4
My work is entirely screen-based and digital.
My work involves significant hands-on, physical, or in-person components (hardware, on-site logistics).

Impact theory

1 = fully left, 5 = fully right. 3 = neither applies, or it varies a lot.

Please answer all highlighted questions before continuing.

Impact theory

G1
I directly work on AI safety problems myself.
My leverage comes from enabling, funding, or multiplying those who do direct work.
G2
The impact of what I do is visible within 1–2 years.
The impact of what I do will likely take a decade or more to become visible.
G3
My work follows a well-established path with reasonably clear success criteria.
I am pioneering a less well-defined path, with significant uncertainty about what success looks like.

Role descriptions (optional)

The next questions fill the role description cards in the career discovery tool. This will be the text that tells someone exploring the field what your role is actually like from the inside. These responses might be the most valuable part of the whole survey for making the tool genuinely useful.

It takes around 5 more minutes. If you're short on time, you can submit now. Your profile data is already saved and is very useful on its own.

What is this role actually like?

Bullet points are very welcome throughout. Quick, honest answers are more useful than polished ones. All questions in this section are optional. Please answer what you can.

Bullet points welcome. Keep it concrete.

Bullet points welcome.

Bullet points welcome.

Bullet points welcome.

Total: 0%

Links, titles, or author names. Bullet points welcome.

Bullet points welcome.

Thank you.

Your response has been recorded. It will be anonymised and averaged with others in similar roles to build the career discovery tool.