home · /programs/fellowship

Neurodivergent Researcher Fellowship

aspirational · funding-contingent

Path: /programs/fellowship.md · v1.0 · 2026.05

The portfolio's neurodivergent-first methodology is currently a method-paper signed by a single author. A fellowship converts that into a structural commitment: it brings neurodivergent researchers into the program as designers, not subjects, on stipends sufficient to support real research time. The fellowship is funded-contingent and listed on the roadmap as Aspirational until the funding tier supporting it lands.

This program exists because methodology that names a population should be built with that population. The neurodivergent-first methodology paper is the warrant; this fellowship is the operational follow-through.

What the fellowship is

A 6-month, paid research fellowship for one or two neurodivergent researchers per cohort.

Each fellow: - Receives a $30,000 stipend over 6 months. - Selects a research focus aligned with the Constitution's amendments (cognitive, human, epistemic, or empirical safety). - Receives mentorship from the founder and the (seated) advisory board. - Co-publishes work under the Constitution's open-licensing framework. - Retains full rights to their work and may publish independently if they prefer. - Is given access to the corpus (under RRL-v1), the apparatus, and the publication infrastructure.

Two cohorts per year. Up to two fellows per cohort. Maximum 4 active fellows at any time.

Eligibility

Open to researchers who:

Self-identify as neurodivergent (autistic, ADHD, dyslexic, schizophrenic, OCD, Tourette's, learning-disabled, or otherwise outside the dominant cognitive profile). No medical documentation required; self-identification is the standard.
Have demonstrable research, engineering, or writing experience relevant to behavioral AI safety. Forms of demonstration include: published papers, shipped software with safety implications, blog posts engaging substantively with adjacent literature, or substantive critiques of existing safety work.
Are not currently employed full-time at a major AI safety lab. (We don't want this to be a sabbatical program for already-funded researchers.)
Can commit ~25 hours/week to the fellowship during the 6 months.

We do not require: - A degree. - An institutional affiliation. - Publications in particular venues. - Citizenship or residency in any particular country. - A particular set of identity intersections beyond "neurodivergent."

Fellows from the global south, formerly housing-insecure researchers, and researchers whose neurodivergence has been a barrier to conventional safety pipelines are explicitly encouraged to apply.

What we offer beyond the stipend

Access to the apparatus. Fellows can build against PROACTIVE, SentinelOS, the corpus, and the published artifacts as if they were core contributors.
Mentorship time. Weekly 60-minute mentor meeting with the founder; quarterly meeting with the advisory board.
Publication infrastructure. Drafts can be hosted on the Constitution's site with their own URL space. Fellows retain copyright.
Cohort. Two fellows per cohort means each fellow has at least one peer. Cohort calls weekly.
Compute budget. $2,000 per fellow for cloud compute or hardware needs (lockfile-pinned via the fiscal sponsor).
Travel budget. $3,000 per fellow for conferences or research visits.

What we ask

A research proposal at application (~1,000 words). What question; why this question for behavioral safety; what apparatus you plan to use or build; what success looks like.
A 30-day check-in in the first month. Honest reflection on whether the fellowship is the right vehicle for the research.
A mid-point review at month 3. Specific deliverables completed or revised.
A final output. Form is flexible: a paper, a tool, a corpus, a methodology amendment, a critique. The form should match the research; we don't impose a single output type.
A reflective essay at the end (~1,500 words). What worked; what didn't; what the fellowship could improve. Published with the fellow's consent.

We don't ask fellows to be neurodivergent ambassadors. We don't ask them to write about their own neurodivergence unless they want to. We don't ask them to validate the methodology paper. We ask them to do their research.

Selection process

Open call. 60-day application window per cohort. Calls published on the site, in disability-research networks, on Mastodon, on LinkedIn, on neurodivergent-researcher mailing lists.
Initial review. Founder reads all applications. Shortlist of 6 produced for cohorts of 1, 8 for cohorts of 2.
Conversation. 45-minute conversation with each shortlisted candidate. Not a "test." A real conversation about the proposed research.
Selection. Final selection by the founder in consultation with the advisory board (when seated). Selection criteria published at /programs/fellowship-selection-criteria.md.
Notification. Selected fellows notified within 30 days of application close. Non-selected applicants receive specific feedback if they request it.

We aim for 50% offer-acceptance rate (i.e., we expect some declines and we plan for them).

What success looks like

Per fellow, in priority order:

The fellow produces a research output the field cites or uses. Citation is a strong signal; usage is a stronger one.
The fellow's experience with the fellowship is positive and they say so publicly (only if they want to).
The fellow continues in the field — independent research, academic role, lab position — with the fellowship as part of their public trajectory.

Per program, after 4 cohorts (2 years):

A coherent body of fellow-produced work on a /programs/fellowship/cohort-N/ directory.
At least two fellows whose work has been cited outside the Constitution.
A methodology paper v2 amended in response to fellow-driven research.
A reputation as a serious neurodivergent-researcher pathway in AI safety.

What this program is not

Not a charity. Fellows are researchers, not beneficiaries. The stipend is compensation for work, not a grant of pity.
Not a diversity check-box. The program produces research outputs that should be evaluable on the same standards as other published work.
Not unlimited. 2-4 fellows per year is the operational ceiling at current funding tiers.
Not aimed at credentialing the Constitution. If a fellow finds their work conflicts with the Constitution's framing, the fellow's work is published with the conflict noted; we do not require fellows to align.

Honest disclosures

This program does not exist as of v1.0 of the roadmap. It is funded-contingent and listed as Aspirational. The first call opens within 60 days of the $480k funding tier being reached, or within 12 months of the $190k tier being reached, whichever comes first. If those tiers are not reached, the program does not exist.

We publish this template so that funders evaluating the $480k tier can see what it is buying. If the tier is funded and the program is launched, this template becomes operational rather than aspirational and migrates to /programs/fellowship-active.md.

Funder messaging note

A funder considering whether to underwrite this program should know:

This is the most distinctive program in the funding ask. There are well-funded AI-safety fellowships; there are not well-funded neurodivergent-first AI-safety fellowships.
The fellowship's outputs accumulate over years; the funder is buying a multi-year cohort, not a single research output.
A funder who underwrites this is named in the fellowship's public materials, with their consent. We do not anonymize funders by default.
A funder who later considers expanding to a longer or larger fellowship is welcome to do so as a co-funder rather than a sole funder.