Detailed research note

CAIS funding and creator-pathway findings

Part of the research through 9 September 2026. This dated note preserves its original findings; later developments are discussed in the synthesis and linked profiles.

CAIS funding and creator-pathway findings

Research cutoff: 2026-09-05. Public-source inquiry. Claims below distinguish organization statements, public tax data, indexed historical source text and inference. No outreach. No evidence obtained of undisclosed editorial orders or of the identity of the screenshot sender described by Sabine Hossenfelder.

Main result

CAIS has a documented public communications pathway connecting a research nonprofit, a separate policy advocacy nonprofit, and creator training/placement. The historical donor trail reaches Good Ventures (Cari Tuna / Dustin Moskovitz, advised by Open Philanthropy, now Coefficient Giving), SVCF, and FLI. The public evidence does not establish who financed a particular 2026 creator offer, current donor concentration, or donor control of particular videos. A historically indexed October 2023 Good Ventures entry is explicitly labelled an exit grant; do not casually call Open Phil/Good Ventures CAIS's current paymaster.

Source registry (URLs are exact; checked 2026-09-05)

S1. https://projects.propublica.org/nonprofits/organizations/881751310 — IRS-derived nonprofit index, financial and officer extracts read. CAIS EIN 88-1751310; 501(c)(3); exempt since Jan 2023. Years 2022-2024; FY2024 filed Nov 17 2025; FY2023 June 9 2025; FY2022 Nov 16 2023. Raw embedded return/PDF unavailable in web parser; XML endpoint returned security check in one alternative retrieval, branch stopped. S2. https://projects.propublica.org/nonprofits/organizations/932442608 — IRS-derived index read. CAIS Action Fund EIN 93-2442608; 501(c)(4); exempt since Oct 2023. FY2024 filed Nov 17 2025; FY2023 Nov 15 2024. Same raw-return access limit. S3. https://www.goodventures.org/our-portfolio/grants-database/page/109/?limit=all — historical funder database text read in search-index result. Live URL redirects to https://goodventures.org/our-giving/ without the historical rows in extracted text. Four CAIS awards below. Treat as indexed historical primary-source evidence, not live database verification. S4. https://goodventures.org/ — live official page read. Good Ventures describes Coefficient Giving (formerly Open Philanthropy) as its primary philanthropic advisor and identifies creators Cari Tuna and Dustin Moskovitz. S5. https://goodventures.org/governance/ — live official page read. Private-foundation board: Cari Tuna (chair), Dustin Moskovitz (secretary/treasurer), Divesh Makan (partner, ICONIQ Capital). S6. https://goodventures.org/our-story/ — live official founder statement read. Good Ventures created in 2011; founders intended to give away vast majority of wealth; Moskovitz's Facebook-to-Asana background described. This is origin/governance evidence, not a particular transfer record. S7. https://www.svcf.org/assets/img/media/SVCF-2023-990-PDC.pdf — official-hosted SVCF public disclosure Form 990 for calendar 2023 read selectively. Prepared Nov 5 2024; generated Nov 7 2024. PDF page 87 (zero-index 86), Schedule I continuation, row 518: CAIS EIN 88-1751310, cash grant $1,455,200, purpose SCIENCES. Relevant page screenshot retrieved. Upstream donor/account not identified in row. S8. https://futureoflife.org/grant-program/2023-grants/ — live official grant-program page read. CAIS $22,000 recommended, general support, 2023; page labels program funds allocated. Recommendation/award disclosure, not bank-level proof. S9. https://safe.ai/about — live official CAIS overview and leadership read. Dan Hendrycks executive/research director; Oliver Zhang managing director; Josue Estrada COO. Research, field-building and advocacy mission explicitly described. Not a complete board register. S10. https://action.safe.ai/ — live official Action Fund identity, approach and priorities read. Nonpartisan advocacy; U.S. AI leadership, compute security, malicious-use prevention, global cooperation. Legal/tax separation stated. S11. https://action.safe.ai/team — live official team page read. Varun Krovi executive director; self-description references Capitol Hill chief-of-staff experience, NIST RMF and CHIPS and Science Act work. Specific causal policy credit is self-reported. S12. https://danhendrycks.com/ — live self-authored professional biography read. Describes Hendrycks as CAIS executive director and advisor to xAI and Scale AI. Terms, compensation and authority not disclosed. S13. https://www.framefellowship.com/ — live official creator-program page read. CAIS participates in Amplifier Track; independent track promises editorial control; program reports Veronica Hylak secured CAIS contract. Past mentor Max Winga identified with Control AI partnerships. Contract terms, amount and outputs not supplied. S14. https://forum.effectivealtruism.org/posts/bQDBB8L9RjPmfGLiB/meet-the-fellows-frame-fellowship-cohort-2-0 — organizer Akshyae Singh announcement, displayed Aug 27 (2026 inferred from current cohort dates/context), read. Sept 7-Nov 15 cohort; names Sara Landi Tortoli (Sara in WonderTech) to work with CAIS translating safety research into public content. Also describes Irreplaceable as originating from CAIS. Planned placement as of cutoff, not completed outcome. S15. https://coefficientgiving.org/research/ai-safety-and-security-need-more-funders/ — official funder strategy article, indexed as published about 11 months before cutoff (exact date not extracted), substantive text read via search result. Explains funding concentration and desirability of politically/geographically appropriate donor identities; says some grantees seek minority funding from one source, creating implicit matching opportunities. Macro-level context only; no proof CAIS uses these mechanisms.

Financial scope (USD; separate entities)

CAIS FY2022: revenue 6,660,060; expenses 816,760; net assets 5,843,300. CAIS FY2023: revenue 16,089,191; contributions 15,594,432; expenses 8,134,449; net assets 8,538,240. CAIS FY2024: revenue 10,235,085; contributions 9,596,669; expenses 7,163,607; net assets 11,609,718. Latest accessible year; source S1. Action Fund FY2023: revenue/contributions 718,870; expenses 139,139; net assets 579,731. Action Fund FY2024: revenue 3,577,153; contributions 3,532,366; expenses 2,394,762; net assets 1,762,122. Latest accessible year; source S2. Do not sum awards into revenue shares without reconciling cash dates, accrual recognition, restrictions and fiscal periods. No complete current donor list was recovered. CAIS funding percentages from 2022-23 do not answer who funds its 2026 communications.

Typed edges

from | to | type | amount | currency | period | source | certainty / limits Good Ventures / Open Phil-advised funding | CAIS | grant_award_general_support | 5160000 | USD | Nov 2022 | S3 | Indexed historical primary text; payment date not established. Good Ventures / Open Phil-advised funding | CAIS | grant_award_philosophy_fellowship_and_NeurIPS_prizes | 1433000 | USD | Feb 2023 | S3 | Indexed historical primary text; restricted purpose, not creator grants. Good Ventures / Open Phil-advised funding | CAIS | grant_award_general_support | 4025729 | USD | Apr 2023 | S3 | Indexed historical primary text. Good Ventures / Open Phil-advised funding | CAIS | grant_award_exit_grant | 1866559 | USD | Oct 2023 | S3 | Indexed historical primary text explicitly calls it exit grant; subsequent relationship not established. Silicon Valley Community Foundation | CAIS EIN 88-1751310 | reported_cash_grant_sciences | 1455200 | USD | CY2023 | S7 | Primary filed row; originating donor unknown; avoid treating SVCF as ultimate donor. Future of Life Institute | CAIS | recommended_general_support_grant | 22000 | USD | 2023 | S8 | Organization disclosure; program marked funds allocated. Coefficient Giving (formerly Open Philanthropy) | Good Ventures | primary_philanthropic_advisor | null | null | current page Sep 5 2026 | S4 | Official role; not a separate additional grant. Cari Tuna / Dustin Moskovitz | Good Ventures | founders_and_governance | null | null | 2011-current | S4-S6 | Foundation wealth origin/roles; not proof every grant came from the same assets. Divesh Makan | Good Ventures | board_member | null | null | current Sep 5 2026 | S5 | Does not establish ICONIQ as donor or controller. Dan Hendrycks | CAIS | executive_research_director | null | null | current Sep 5 2026 | S9,S12 | Official/self biography. Dan Hendrycks | CAIS Action Fund | president | null | null | FY2024 | S2 | IRS-derived extract; compensation $0 at AF in extract. Varun Krovi | CAIS Action Fund | executive_director | null | null | FY2024/current | S2,S11 | Historical and current agree. Dan Hendrycks | xAI; Scale AI | professional_advisor | null | null | current biography Sep 5 2026 | S12 | Amount, role scope and decision power unknown. Frame Fellowship | CAIS | amplifier_track_host_partnership | null | null | cohort 2 Sept 7-Nov 15 2026 planned | S13,S14 | Publicly announced hosting, not proof of specific paid output. CAIS | Veronica Hylak | creator_contract_reported_by_program | null | null | before Sep 5 2026; exact date unknown | S13 | Contract existence is program assertion; no executed contract or amount read. Frame Fellowship / CAIS | Sara Landi Tortoli | announced_creator_placement | null | null | Sept 7-Nov 15 2026 planned | S14 | Future work at cutoff; not an accomplished campaign. Max Winga / Control AI | Frame Fellowship | past_mentorship | null | null | before cutoff, exact dates unknown | S13 | People/program connection; no CAIS-ControlAI money edge inferred.

Four historical Good Ventures award entries total $12,485,288, but that is an award inventory, not a reconciled received-funding total. Open Philanthropy recommendations and Good Ventures awards must not be double counted.

Story assessment and rival explanations

Strongest defensible finding: CAIS combines technical credibility, public advocacy and access to a creator-training/placement network; it is not just a laboratory. The Action Fund is openly concerned with national-security/leadership policy, so describing the entire complex simply as anti-AI obscures stated objectives. Industry advisory roles provide access and possible conflicts worth scrutinizing, but do not prove an industry secretly controls CAIS.

A plausible influence story is that large philanthropic commitments help build institutions and trained communicators who shape public and policy agendas. The rival explanation is mission-aligned research and risk education with independent or negotiated editorial arrangements. Frame explicitly offers both independent and organization-amplifier modes, so one blanket independence claim for all fellows would also be wrong. Funding, access, planned cooperation, and editorial control require different evidence.

The historical exit-grant label is discriminating evidence against assuming continuous Good Ventures control. SVCF is a material funding intermediary whose ultimate contributor cannot be read from this row. The FLI connection is real but small in this specific 2023 grant disclosure. No Constellation-to-CAIS funding edge was established; shared ecosystem references and compute acknowledgments are insufficient.

Gaps and finite next steps

The screenshot sender/offer is unidentified; Sabine's own declined sponsor is separate and unnamed. No specific current CAIS creator-grant application terms, total budget, creator payment, disclosure clause, or editorial veto was found. A public screenshot or grant agreement would discriminate better than more generic organization searches. Complete current boards and inter-entity transfers require full relevant 990 schedules or public regulator records; index officer names are not a complete board roster. CAIS 2024 impact-report landing page https://safe.ai/work/impact-report/2024 links https://odw.dev/cais/CAIS_impact_report_2024.pdf but PDF retrieval failed (web 502; one PowerShell substitute SSL failure); no report content or audited financial statements were certified read. CAIS raw 990 page/PDF parser and XML security-check failure are access limitations, not suspicious findings. Secondary leads not promoted: FTX/FTX Future Fund startup funding and bankruptcy subpoena; Schmidt-related grants; OpenAI 2023 990 grants. These require primary records. Do not let a secondary database's apparent FY2025 label turn historical Good Ventures awards into new 2025 funding.

No relevant memory hit was found in the single registry search; no memory fact was used. No outreach or publication. This note is the workstream handoff, not a completed forensic audit.