ControlAI upstream funding and resource follow-up

Observed 2026-09-05. Bounded public-source inquiry in six groups: direct fundraising disclosures; platform grant records; US nonprofit identity/filing; named grantmaker searches; Conjecture resources/transition; independence and adjacent institutional links. No outreach, private access, paid access or publication. UK accounts already examined by parent were not reopened.

Result

This pass found a new US legal-entity/financial lead and a concrete prospective grant application, but did not identify an independently verified donor-to-ControlAI transfer or binding donor conditions. The strongest new distinction is between money the US entity reportedly received in 2025, money requested in a July 2026 application, and the much larger hypothetical budgets advertised in an April fundraising pitch.

The relationship to Conjecture is well documented through people and collaborative work. It remains unproved as an actual transfer of cash, assets, intellectual property, office subsidy, or payroll support in this inquiry. Donations to Talos and investment in Conjecture must not be counted as donations to ControlAI.

1. ControlAI's direct fundraising appeal

Source: https://controlai.org/blog/prevent-asi-plan Title: Preventing extinction from ASI on a $50M yearly budget. By Andrea Miotti, Alex Amadori and Gabriel Alfour; April 20, 2026. Opened/read.

It estimates $50m/year would give ControlAI a chance at achieving an international ASI prohibition, and discusses $500m-$1bn budgets as possible further expansion. These are requested/hypothetical scales, not actual revenue, realized gifts or approved spending. It invites major philanthropic support and gifts of $100k or more, saying it is not presently set up to accept small donations. It describes itself as a US 501(c)(4) and UK nonprofit that is not a registered charity, and says gifts are not tax deductible.

Source locations: opening lines 7-15; fundraising section 216-221; media collaboration 199-203. It says 14 creator collaborations included paid and free arrangements. No donor list, donation agreement, board seat for a funder, approval right or binding funding condition was found here. Do not extrapolate the named creator work to every video or every campaign.

2. Application, not an award

Source: https://app.grantmaking.ai/orgs/f723327d-43a4-4e02-ae12-5f68bb6f086f Opened/read. Platform-hosted organization and application record; the surrounding organizational summary is edited by grantmaking.ai and includes stale/conflated descriptions, so use the application and status fields narrowly.

Active application: Scaling ControlAI's parliamentary outreach across the G7 (excl. US), Applied July 13, 2026; $50,000 requested minimum, up to $1,000,000. Organization Grants Received field says no grants recorded, which describes this database, not all real funding. Text says a grant would be earmarked for the program. This is prospective wording, not proof of an executed restricted-grant agreement.

Minimum proposal: $45,000 for one outreach officer over 6-8 months, plus $5,000 travel/events; expected 20-40 additional legislators briefed. $1m proposal: $800,000 up to eight FTEs, $40,000 events, $25,000 software/materials, $25,000 travel/recruiting, $110,000 operations/contingency, over 12 months. Expected 300-600 additional lawmakers briefed and another 20-plus-signatory cross-party campaign.

Scope inconsistency: heading and repeated passages exclude the US and include Brussels, but one country-selection sentence includes existing US teams. Keep this unresolved; do not silently harmonize it. Individual project page failed to open.

Fund's own platform record: https://app.grantmaking.ai/funds/4597b683-2776-4731-8598-627189912dc7 . Opened/read Grants Made list; no ControlAI grant shown. No actual award or paid amount established. Site-wide distributed totals are not ControlAI receipts.

3. Newly identifiable US entity and 2025 financial lead

Source: https://www.causeiq.com/organizations/control-ai-us,333361748/ Opened/read 2026-09-05. IRS-derived secondary source, not the original return.

Entity: Control AI US; EIN 33-3361748; 501(c)(4); Washington, DC; corporation; fiscal year ends December. Personnel names Andrea Miotti, Connor Leahy and Sophie Toura match the campaign organization, and the site identifies controlai.org. No c3 entity identified in this pass.

Reported FY2025: contributions $136,570; other income $377; total revenue $136,947; expenses $93,682; total assets $69,178. Listed return: Form 990-EZ, fiscal year December2025, received by IRS June23,2026; platform added it August6,2026. These values remain aggregator-reported. Do not use the US entity's limited-year finances as the entire global ControlAI budget or derive donor concentration from them.

Original return link: https://www.causeiq.com/organizations/view_990/333361748/f9c3cde1c233708dd35572ad3ab7814b . Web fetch failed; one direct public GET returned403. No PDF saved. ProPublica entity https://projects.propublica.org/nonprofits/organizations/333361748 failed; public API alternative https://projects.propublica.org/nonprofits/api/v2/organizations/333361748.json unavailable to web. Recovery stopped. No original Schedule O, individual donors, related-party transfers, or specific restrictions established.

4. Direct donor database searches

Targeted public searches for ControlAI and/or Secure Future Research in Survival and Flourishing Fund, Coefficient Giving/Open Philanthropy and Manifund did not identify a grant specifically awarded to ControlAI. This is search non-discovery, not an exhaustive database audit or proof no funding occurred.

Manifund results included grants to other organizations and creators mentioning ControlAI, e.g. Apart Research's collaboration and independent creator Tom Bibby's references to ControlAI. These were not counted as ControlAI receipts. No indirect route was turned into a direct funding claim.

5. Conjecture resources and personnel transition

Source: https://www.ettf.land/p/conjecture-a-retrospective Connor Leahy, March26,2026. Opened/read.

Leahy describes Conjecture's investor-backed founding, technical research, product pivot and failures, and announces joining ControlAI US. He credits an initial investor by first name, Nat, but supplies no new ControlAI donor identity or transfer. This supports a personnel/intellectual trajectory, not a financial merger, assignment of assets or continuing Conjecture-funded payroll.

Source: https://www.ettf.land/p/conjecture-a-retrospective/comments Opened/read. Leahy calls the field money-constrained and solicits people able to mobilize over $100k. A reader asks whether ControlAI sought/rejected OpenPhil funding, but no answer appears. That question is NOT evidence of either an application or rejection. Claims about controlled opposition are Leahy's polemical assessment, not established facts of this investigation. No new funded amount found.

New corroboration for previous Behrens findings: https://www.talosnetwork.org/perspectives/alumni-newsletter-october-2025 . Opened/read, dated October14,2025, author David Conrad. Latest Publications explicitly identifies Behrens as Summer2022 alumna and reports Conjecture/ControlAI coauthorship. This upgrades the earlier indexed-only Talos cohort observation to a read primary publisher source; no fellowship payment or donor earmark follows from it.

6. Independence, resources and adjacent coalition lead

Source: https://controlai.b-cdn.net/2025-controlai-impact-report-v2.pdf Primary impact report, December2024-January2026; opened/read, 11pages. Reports team fewer than15 and a small budget, but no exact consolidated financial statement/donor roster. Describes policy/media/creator outreach and measurable engagement; not funding obligations. The pitch and impact report together disclose uses and ambitions much more clearly than actual upstream funders.

Canadian registration lead: https://lobbycanada.gc.ca/app/secure/ocl/lrs/do/vwRg?cno=382906&regId=976385 . Search-indexed French disclosure says Secure Future Research activities are not controlled/directed by another interested entity in that engagement; opening returned403. Not used as an independently read or universal funding-independence conclusion. Parent access work may establish the precise field and scope.

Optional coalition lead, with strict limit: https://www.cbpai.org/deal-of-the-century-rome-convening-2026 . Opened/read. CBPAI's own archived event page lists ControlAI/AItreaty among coalition organizations and says CBPAI was seed-funded by SFF. This is coalition-side attribution, not a direct ControlAI acknowledgement or a SFF-to-ControlAI grant. It does not establish payment to participants. Archive notice says the Rome event was merged into Washington September15-16,2026; as of this inquiry that event is future, not demonstrated access already achieved. Do not infer attendance from its proposed invitee categories. No expansion into its donor network undertaken.

Typed edges

Unresolved questions that matter

Actual ControlAI funder names and amounts; restrictions, reporting terms, termination/clawback or approval rights; any Conjecture cash/assets/payroll supplied to the nonprofit; transfers between Secure Future Research and Control AI US; US original return and related-party explanations; whether the July application was later decided; the application geographic inconsistency. These remain gaps. Neither transparent public strategy nor donor opacity by itself proves donor command or hidden coordination.