Shared Screening for AI Safety Hiring of Generalists

Barcelona, Catalonia, Spain

Reduce duplicative hiring processes in AI safety organisations by sharing common screening processes for generalist roles.

Project Details

Aim: Reduce duplicative hiring processes in AI safety organisations for generalist roles.

Note: generalist here refers to any work outside of research and policy, for which sufficient talent pipelines already exist. Generalist roles may refer to finance, legal, operations, marketing, communications, software engineering, etc.

Organisations tend to screen for similar traits in candidates, such as prior engagement with AI safety, reasoning / logic questions, engagement with EA and adjacent communities, and mission alignment. Work tests are also likely to be duplicative.

I'd like to explore a proof of concept for interested organisations to use a standardised approach to pool and screen for generalist positions, with shortlisted candidates completing a work test specific to their area of expertise for which all participating hiring organisations can see the results.

Phase 1 - Research: Consult with 8-10 AI safety organisations that plan to expand significantly over the next few years to find out where their hiring processes overlap and what signals they would trust from a shared process. This research would be anonymised and shared on the EA forum as a report on duplicative hiring and important signals for AI safety generalist roles.

Phase 2 - Build a prototype with 50 candidates in a searchable database that includes verified AI safety engagement (e.g. Bluedot course completion), reasoning questions, and a short recorded video on their interest in AI safety. It would also include questions specific to their field of expertise. An up-to-date database of pre-vetted candidates for generalist roles would not only speed up hiring for full-time roles, but also enable orgs to easily find candidates for contract/freelance projects.

Phase 3 - Pilot shared work tests: candidates that get shortlisted by a participating org complete a work test relevant to their field, of which the results are shared with all participating orgs (with the candidate's consent). Work tests would be generated with the help of experts in the field, both inside and outside of AI safety. The pilot would focus on work tests for operations roles.

I will do the research, design and build the prototype of the database, and pilot shared work tests (possibly using Truffle or similar). My background is in product design and front-end development. I will use AI tooling to speed up development, and contacts within AI safety and with experience in operations and hiring for guidance along the way.

The concrete output will be:

a report on duplicative hiring for generalist roles within AI safety
a publicly accessible database of people interesting in working in AI safety with fields directly relevant to their field of work
work test results that will be shared amongst participating organisations for operations roles

Theory of Impact

Hiring in AI safety is slow and costly: rounds typically take 3–6 months, require significant time from senior staff/researchers, and often go through referrals of people already in the field. This pushes talented applicants out of the field; mid-to-senior talent cannot afford to volunteer or apply indefinitely for full-time roles. It also keeps AI safety insular and slow to learn from experienced external professionals. Meanwhile, AI safety is expected to get more funding than it can handle in the next few years. To make that money impactful, it needs to attract and be able to hire exactly those people who bring proven operational, communications, and engineering skills from outside the field.

People

Emily Marais

Team Member

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$12,000

Ideal

$18,500

What this ask is

Aim: Reduce duplicative hiring processes in AI safety organisations for generalist roles.

The concrete output will be:

a report on duplicative hiring for generalist roles within AI safety
a publicly accessible database of people interesting in working in AI safety with fields directly relevant to their field of work
work test results that will be shared amongst participating organisations for operations roles

How the money will be spent

Minimum scope (phases 1-2: org consultations + public research report, prototype candidate database):

$10,600 (part-time at 2,600 GBP/mo for 3 months)
~$800 for Claude Code subscriptions and interview recording tooling
5% contingency
Total: $11,970

Ideal scope: (adds phase 3: shared work-test pilot: further research into work tests, consultations with experienced operations professionals, test design, 10–15 paid candidate work tests shared across participating orgs)

+$4,550 (part-time at 2,600 GBP/mo for 1.5-2 months)
+$1,500 paid candidate work tests (12 × $125)
+$200 extended tooling
+5% contingency
Total: $18,500

Discussion

MMark AikenEndorsement15d(edited)

I endorse this idea / proof of concept.

Existing AI safety recruitment processes have several stages, with the early stages typically testing similar things around background, prior engagement with AI, alignment / context etc. There's duplication and repetition for recruiting organisations and candidates at this stage, so it would be interesting to test whether it's feasible to establish some kind of pre-qualification process. There are already AI safety rosters of course, but these are mostly either 'opt-in' without screening, or screened on the basis of completing a training course / programme. It would be interesting to test whether the work test approach was any more or less successful in identifying suitable candidates. It's also a reasonably common model in other domains.

I see that Emily worked on the AISafety.com website update.