AI Safety Support

Theory of Change

AISS aimed to reduce AI existential risk by supporting the people working on it -- not by doing research itself, but by helping early-career, independent, and transitioning researchers become effective. In JJ Hepburn's words: "I think too many people feel held back from doing a project like thing on their own... Just do a project. Drop the constraints."

The causal chain: provide career coaching, health coaching, community (AI Alignment Slack), feedback events, and mentorship matching --> more people enter and persist in AI safety work --> more total research output --> lower AI risk. Separately, AISS's fiscal sponsorship function (later spun out as Ashgro Inc) provided the legal and administrative infrastructure for programs like MATS, Astra Fellowship, and ARENA to receive and deploy grant funding.

These were two distinct theories of change housed under one organizational umbrella. The first (support services) died in July 2023. The second (fiscal sponsorship) thrived and now processes $4.2M/year as Ashgro Inc.

What They Do

AISS support services (shut down July 2023): Career coaching (1-on-1 calls for career changers, independents, students), health coaching (free to AI safety workers via Shay Gestal), the AI Alignment Slack workspace, research feedback events, mentorship matching, resource library, and newsletter. All services were free to recipients. Five published testimonials exist from software engineers, PhD students, and career changers.

AISS Ltd fiscal sponsorship (ongoing): The Australian legal entity (ACNC charity) continues to receive Coefficient Giving grants for programs it hosts. $7.85M total for MATS ($6.6M across 5 grants), Astra Fellowship ($617K), ARENA ($146K), Owain Evans research ($444K), and Epoch ($42K). Most recent CG grant: October 2024, $710K for MATS extension. MATS website still credits "AI Safety Support" for $15K per-fellow stipends.

Ashgro Inc (successor entity, active): US 501(c)(3) founded 2022 by JJ Hepburn. Model A fiscal sponsorship (comprehensive -- project leads become Ashgro employees). 5-10% fee on project revenue. Sponsors at least 10 AI safety projects including Timaeus, Apart Research, Robert Miles/aisafety.info, AI Safety Camp, AFFINE, CORAL, Catalyze Impact, Medronho, and Alignment Ecosystem Development. Revenue grew from $71K (2022) to $4.2M (2024). Three employees, 50 volunteers.

AI Safety Quest (grassroots successor): Volunteer-based org that picked up AISS's community support role. Offers "Navigation Calls" (1-on-1 advising), mentorship matching, and group learning coordination. Runs through AED, fiscally sponsored by Ashgro. Unfunded.

Key People

JJ Hepburn -- Co-founder/ED of AISS, CEO of Ashgro. ANU graduate with background in electrical contracting (HighSpecs) and data services (CanvasU). Entered AI safety via AISC3/AISC4. Takes $0 cash salary from Ashgro ($44,675 "other compensation" in 2024). Founded both entities. The common thread across AISS and Ashgro is JJ's focus on removing barriers for people who want to work on AI safety.

Linda Linsefors -- Co-founder of AISS and initiator of AI Safety Camp. PhD in theoretical physics. Returned to AISC research coordination in 2023 after stepping back from AISS. Authored detailed community resource guides.

Robert Miles -- AISS advisor. AI safety YouTuber (145K subscribers). Now fiscally sponsored by Ashgro ($122K LTFF grant for YouTube/aisafety.info). His involvement bridges AISS's original community-building mission with Ashgro's current infrastructure role.

Team was ~4 people at peak (JJ, Rachel Williams as COO, Frances Lorenz ops, Shay Gestal health coach). Ashgro is similarly small: 3 employees, hiring an operations associate in late 2025.

Money and Incentives

AISS's own funding was thin:

Survival and Flourishing Fund: $200,000
FTX Future Fund: $200,000 (funder collapsed November 2022)
EA Funds: $25,000
Total for support services: ~$425,000

Flow-through grants (fiscal sponsorship for other programs):

$7,851,528 in Coefficient Giving grants to "AI Safety Support" for MATS, Astra, ARENA, Owain Evans, Epoch
Ratio: for every $1 AISS received for its own mission, $18 passed through for other programs

Ashgro Inc financials (990 data):

Year	Revenue	Assets	Expenses	Salaries
2022	$71K	$61K	$15K	$12K
2023	$2.1M	$1.66M	$528K	$188K
2024	$4.2M	$3.36M	~$2.5M (est)	$879K

Revenue is 98.7% contributions (pass-through grants). Fee income is embedded in the 0.6% "program services" line ($25K in 2024) plus potentially a portion of contributions. The 5-10% fiscal sponsorship fee on $4.2M in grants would generate $210K-$420K in operational revenue -- a sustainable model.

Business model: AISS's support services had no revenue model (free services, dependent on grants). Ashgro's fiscal sponsorship has a built-in revenue model (percentage fee on grants processed). The part with a revenue model survived; the part without one died.

The LTFF/donor signaling failure: A potential individual donor was willing to fund AISS for a full year but hesitated because LTFF hadn't decided on AISS's application. The donor assumed LTFF's indecision signaled concerns about AISS. In reality, LTFF was itself funding-constrained. This cascading information failure likely contributed to AISS's death.

Incentive structure: Ashgro's incentives are reasonably aligned. It charges fees to provide administrative services, allowing safety projects to focus on safety. JJ takes near-zero salary. The main risk is that as fiscal sponsor, Ashgro's revenue grows when it sponsors more and larger projects, creating incentive to prioritize growth over quality control -- but at current scale (3 employees, $4.2M), this seems distant.

What Others Say

No substantive criticism exists. Despite extensive searching, no one has publicly criticized AISS's approach, effectiveness, or theory of change. The org was too small and too well-liked to attract critics. The strongest implicit criticism is systemic: the AI safety funding ecosystem was not designed to sustain community infrastructure work.

JJ on the shutdown: "A lack of funding is a part of the decision but is not the only factor." In comments: he reached a point where even with full funding he wouldn't want to run AISS through another funding cycle. His experience with EA Funds/LTFF was "a significant part" of the shutdown decision. He didn't want to be an applicant again.

LTFF on community infrastructure: Fund managers reviewing Alignment Ecosystem Development (an Ashgro-sponsored community project) were split: one "strongly supported the grant," while two others "were not convinced that it met the fund's high funding bar." Concerns included "spreading themselves too thin" and that newcomers "might become unimpressed with the field" from community infrastructure work. The grant passed on one enthusiastic vote overriding tepid consensus.

Ashgro clients: "Ashgro has made aisafety.info possible by shielding us from bureaucracy" (Plex). "JJ and the team are very responsive, deeply knowledgeable about nonprofit operations" (Jesse Hoogland, Timaeus).

What's Absent

No impact metrics from AISS. No published data on sessions conducted, users served, Slack community size, or outcomes.
No financial transparency from AISS Ltd. Australian entity has no US 990 equivalent accessible. Unknown if AISS collected fiscal sponsorship fees on $7.85M flow-through.
No public interviews or podcasts with JJ Hepburn. The founder's views are known only from three short forum posts.
No information on what happened to the AI Alignment Slack after the shutdown.
No conflict-of-interest or governance policies published by either entity, despite Ashgro processing $4.2M and having a board member (Gaidashenko) who also works at a sponsored project (FAR.AI).