{"path":"research/sff-2026-and-fiscal-sponsorship-path.md","content":"# SFF 2026 and the Fiscal-Sponsorship Path\n\n*Research document — April 16, 2026*\n*Status: Historical April research doc, live-updated July 2026 for the HSEE / fiscal-sponsor / speculation path.*\n\n> **Factual corrections (verified July 6-9, 2026 against SFF's published pages and the application refresh):** the HSEE theme-round supplemental is a short web form (not a long attachment); HSEE recommendations are announced **end of November 2026** with disbursement **by end of March 2027** (the September/end-of-year timeline below applied to the Main round); the same project may **not** apply to multiple rounds (the Main+HSEE stacking idea below is wrong); and late applications forward only to the *next announced* round (SFF-2027, roughly 9–12 months later), so the auto-forward clause is a safety net rather than a strategy. Speculation grants should be treated as the round-entry / expedited-attention gate, **not** as a ~95% approval-rate claim; the 500-character pitch and speculator visibility are the real levers.\n\n---\n\n## Why This Doc Exists\n\nThe earlier [governance and financing research](governance-and-financing.md) mapped Swedish org structures, civic-tech grants, and EA/rationalist adjacencies at a high level, but it was written before the 2026 Survival and Flourishing Fund round opened and before the Swedish x-risk fiscal-sponsorship pattern became concrete. This doc addresses three specific things that earlier research left implicit:\n\n1. **What SFF actually is and why it is the strongest near-term funding fit for Deliberus.**\n2. **Why fiscal sponsorship — not incorporation — is the practical unlock for a Swedish solo-led project applying to SFF.**\n3. **The one publicly-documented Swedish x-risk project that has navigated exactly this path, as a template to learn from.**\n\nIt is meant to be read alongside `governance-and-financing.md` rather than to replace it.\n\n---\n\n## 1. What SFF Is\n\nThe [Survival and Flourishing Fund](https://survivalandflourishing.fund/) (SFF) is the grant-making vehicle funded directly by Jaan Tallinn — Skype co-founder, co-founder of the [Future of Life Institute](https://futureoflife.org/), co-founder of the [Centre for the Study of Existential Risk](https://www.cser.ac.uk/), and one of the most structurally significant donors in the x-risk and AI-safety ecosystem. SFF exists to fund work that, in the donor's framing, contributes to humanity's survival and flourishing during the window in which advanced AI and other transformative technologies become reality.\n\nTwo things make SFF distinctive among AI-safety-adjacent funders:\n\n- **It uses the \"S-Process\" recommender model** rather than a traditional program-officer model. In the current public description, funders agree on four to twelve recommenders who read applications, work through the recommendation app over a sequence of meetings, and produce recommended grant amounts while funders retain final say. This means applications are not weighted toward a single reviewer's priors, and tactically, a proposal that lands well with even a subset of the panel can still receive substantial funding.\n- **It pairs S-Process rounds with \"Speculation Grants\"** — a required eligibility / expedited-funding gate for consideration in an S-Process round. The live application strategy should treat speculation as the near-term route into SFF attention and possible money, not as a guaranteed or nearly-guaranteed approval statistic.\n\n**Recipient-side mechanics of a Speculation Grant** (verified against SFF's speculation-grants page, 2026-07-30): the speculation application doubles as the request for consideration in the next S-Process round, and the grant is evaluated retroactively there — a speculation grant of $Y \"is counted against the total $Z recommendation\" of that round (the grantee is guaranteed at least $Y; the round tops up only if $Z > $Y). The speculator's own budget settles on that retroactive evaluation (\"impact futures\"), so the practical expectation on a recipient is progress the next round's recommenders can recognize — SFF publishes no formal grantee-reporting regime; funds administration runs through the fiscal sponsor (EA Sweden), and the default openness obligation (§6: open-source / CC-BY / MIT+Apache) applies. Nothing in the terms precludes concurrent grants from other funders; ecosystem norm is disclosure of received/pending funding in each application and no double-funded budget lines, and within SFF itself the count-against mechanism is the no-double-dip rule. Pool scale from the same check: **~40 Speculators, ~$20MM combined budget (up from $4MM at launch), individual budgets typically $400–500k, each individually free to reach out and fund** — roughly double the ~18-speculator figure in earlier notes, which materially improves the odds that at least one speculator picks up a given application, and means a $60K ask fits comfortably inside a single speculator's budget with no pooling required. SFF publishes no application-volume or per-speculator pickup-rate data, so any pickup-odds estimate remains reasoned rather than measured. **Reasoned estimate (2026-07-30, for planning only): ~20–35% that at least one speculator funds this application, strictly conditional on the application actually reaching SFF** — which is itself gated on the EA Sweden fiscal-sponsorship confirmation, not on anything about the project. Upward pressure: ~40 independent decision-makers each needing only one yes, a $60K ask that fits inside a single budget with no pooling, a live system with public research and shipped self-measurement instruments (the evaluator-capacity argument below), and an epistemics/coordination thesis that several speculators explicitly fund. Downward pressure: the panel is adversarial-by-default on software tools (TsviBT's stated skepticism), the demand-side null is the application's own confessed weakness, the 500-character pitch is the entire first impression, and a project with no institutional affiliation and no prior grant history has no warm channel into the pool. Treat the number as a planning input for how many parallel applications the slate needs, never as a forecast to defend.\n\n### 1.1 S-Process Mechanics and Deliberus Analogy (2026-07-09)\n\nAndrew Critch's \"S-process funding\" presentation is worth understanding as more than grant-administration trivia. The core mechanism is not \"take votes\" or \"pool money.\" It asks recommenders to represent their judgments as marginal value functions over applications, then runs counterfactual delegation simulations so funders can see how money would flow under different trust/weighting assumptions. Recommenders update their curves through discussion; funders can also express how much they want to defer to each recommender. The system is built to reduce three familiar funding pathologies: bystander effects, pile-ons, and arbitrary ordering effects from who moves first.\n\nThe important design lesson for Deliberus is that S-process treats conversation and computation as complementary, not interchangeable. The app gives the group a manipulable representation of judgments and disagreements; the meetings do the interpretive work. In the talk, Critch explicitly pushes against replacing trust and discourse with numbers alone. A public donation-log description of S-process rounds says recommenders specify marginal utility functions, adjust them through discussion, and funders specify/adjust utility functions for deferring to recommenders. Lightcone's old S-Process Developer role describes the product work as UI plus mathematical formulas for combined decision-making, which is exactly the \"structured representation + human deliberation\" pattern.\n\nThis is a close cousin to Deliberus, but not the same thing:\n\n- **S-process** structures how evaluators allocate money across applicants, using marginal value curves, delegation weights, disagreement matrices, and repeated simulations.\n- **Deliberus** structures the reasoning content beneath a decision: claims, provenance, standing challenges, arguments, residues, hinges, and where a system may be distorting the discussion.\n\nThe strongest application-level implication is that Deliberus can be framed as infrastructure for the layer S-process still leaves mostly social: what claims were made during discourse, which claims deserve follow-up, what evidence supports them, where disagreement actually lives, and whether a discussion is resolving disagreement or merely moving numbers. That does **not** mean Deliberus should claim to be an allocation algorithm. It means Deliberus is a plausible companion layer for the same class of high-stakes, multi-evaluator decisions.\n\n### 1.2 Manifund / Surplus Ecosystem Read (2026-07-09)\n\nThe July 2026 Austin Chen / Oli Habryka conversation and the Surplus launch materially change how to read Manifund. Manifund is not just a donation page or a small-grant form; it is part of an active attempt to build AI-safety public-goods infrastructure: regranting, better funding platforms, and incubated software projects. Manifund can grant to registered charities and individuals, with for-profits potentially eligible after due diligence, and its US 501(c)(3) structure makes project pages a donation surface as well as a regrantor-discovery surface.\n\nSurplus is the sharper strategic fork. Its launch explicitly names three categories that overlap Deliberus: **AI for epistemics and coordination**, public-facing explanatory websites, and community infrastructure including funding platforms, open-source S-Process tooling, and AI-powered grantmaking/review. Its offer is startup-shaped rather than grant-shaped: a three-month SF program, $100k investment via SAFE at a $2m post-money cap, mentorship, office space, and demo day. A SAFE is not a grant; it is a future-equity instrument. In rough terms, $100k at a $2m post-money cap implies selling about 5% if it converts at that cap, subject to the actual legal terms.\n\nThe consequence is **not** \"apply to every adjacent thing immediately.\" Surplus makes Deliberus more legible as a software/infrastructure bet, but also pulls toward a for-profit, SF/incubator, demo-day path. That path is not logically incompatible with grants — nonprofits can buy from or invest in for-profits, and public-good startups can receive philanthropic support — but it changes the center of gravity: investor-style iteration, ownership dilution, and a stronger need to show product pull. The grant-funded public-infrastructure path keeps ownership/control cleaner and fits the SFF/LTFF framing more naturally.\n\nThe practical rule that survives both paths is **artifact-first**. Manifund/Surplus outreach should not lead with vision alone. It should follow shipped, inspectable evidence: the residue map, the disagreement-preservation benchmark, and a dogfood/friction log that shows Deliberus teaching the funder something about the very reasoning/review process their ecosystem is trying to improve. Until those artifacts are crisp, more dogfooding and iteration is higher-leverage than high-salience outreach.\n\nSFF rounds are funder-identical (Jaan Tallinn) but theme-split. The 2026 rounds published on [survivalandflourishing.fund/2026/application](https://survivalandflourishing.fund/2026/application):\n\n| Track | Budget | Deadline (2026) | Focus |\n|---|---|---|---|\n| **Main Track** | $10–20M | **April 22** | Broad x-risk / civilizational resilience |\n| **Freedom Track** | $2–4M | April 22 | AI use that strengthens freedom, prevents concentration of authority, defends free speech / privacy / association |\n| **Fairness Track** | $2–4M | April 22 | AI that empowers the disempowered, anti-monopoly, fair access |\n| Climate | $2–4M | June 10 | |\n| Animal Welfare | $2–4M | June 24 | |\n| **HSEE (Human Self-Enhancement & Empowerment)** | $2–4M | **July 8** | Tools and infrastructure that augment human cognition, judgment, and agency |\n\nTotal round budget: **$20–40M across all tracks.** The April working model treated Main + one theme round as stackable and assumed a 2026 disbursement timeline. That is now superseded by the July 2026 correction above: the same project may not apply to multiple 2026 rounds, HSEE recommendations are announced at the end of November 2026, HSEE disbursement is by the end of March 2027, and late applications forward only to the next announced S-Process round. For the live Deliberus plan, HSEE + speculation/late routing is the relevant path, not Main+HSEE stacking.\n\n## 2. Why SFF Fits Deliberus Unusually Well\n\nThe [80,000 Hours AI-enhanced decision-making research](80000-hours-ai-decision-making.md) already establishes that Deliberus sits exactly inside the \"highly neglected\" gap 80,000 Hours named. SFF is the first-order funding consequence of that gap: the same donor-network ecosystem reads those 80K Hours problem profiles, the same Jaan Tallinn co-founded FLI, and the same x-risk framing that justifies 80K Hours' prioritization justifies SFF's. If Deliberus is the \"reasoning substrate\" underneath AI-enhanced decision-making (as argued in [civilizational-vision.md](../civilizational-vision.md)), then SFF is the most thesis-aligned funding source currently known to the project.\n\nSpecific track fits:\n\n- **Main Track ($10–20M, April 22)** was the default April entry because Deliberus's civilizational pitch — structured reasoning infrastructure, argument graphs, bridging signals, the convergence thesis as alignment substrate — reads as broad x-risk relevance without needing a theme-specific framing. That path is now historical for 2026, not live.\n\n- **HSEE Track ($2–4M, July 8)** became the live 2026 target and is the stronger thematic match. \"Human Self-Enhancement & Empowerment\" is almost a direct paraphrase of the Deliberus mission as articulated in [civilizational-vision.md](../civilizational-vision.md) and [analysis-and-attunement.md](../analysis-and-attunement.md): augmenting human reasoning without replacing it, making the reasoning layer inspectable, letting users attune to perspectives they couldn't previously reach.\n\n- **Main + HSEE stacking is rejected** by the July 2026 factual refresh. Do not revive this as a strategy for the same project unless SFF's rules change in a later year.\n\n- **Freedom and Fairness tracks** are less direct fits. Deliberus touches both (freedom from concentrated authority via transparent reasoning; fairness via making argument quality legible across groups) but applying to one of those instead of HSEE would likely be a worse-fit framing.\n\n## 3. The Fiscal-Sponsorship Unlock\n\nSFF's eligibility rules require applicants to be either a non-profit with existing charity status, a non-profit with a fiscal sponsor, or a for-profit incorporated in the US / UK / Canada / Australia (with \"select cases\" elsewhere). Individuals without a sponsor cannot apply. This is the *only* structural blocker between Deliberus-as-project and an SFF application.\n\nThe earlier [governance and financing research](governance-and-financing.md) framed the Swedish organizational paths as a staged decision: enskild firma → ideell förening or aktiebolag → eventual operating entity. That framing was correct for self-financed bootstrapping but it misreads the SFF-eligibility question: **SFF does not require that the project itself have charity status. It requires that the project have a fiscal sponsor who does.**\n\nA fiscal sponsor is a registered non-profit that receives and administers grant funds on behalf of an un-incorporated project, typically in exchange for a small administrative fee (commonly 5–10% of grant value). The project doesn't have to be legally part of the sponsor, and the sponsor doesn't have to share the project's mission beyond basic compatibility — the sponsor's role is effectively a charitable fiscal host. This is a standard mechanism in the US EA ecosystem (Rethink Priorities, Centre for Effective Altruism, Giving What We Can, Rethink Charity all sponsor affiliated projects) and it exists in Sweden too.\n\nThe structural implication for Deliberus is that the question \"should we incorporate as an ideell förening before applying?\" is the wrong question. The right question is \"which Swedish non-profit can serve as fiscal sponsor for this application, and what do they need from us?\" Finding an existing sponsor collapses the 3–6 month incorporation timeline into a conversation and a memorandum of agreement, and keeps Deliberus's operational form open for later decision (when revenue models are clearer, when team size is clearer, when the right long-term structure is actually knowable).\n\n## 4. Public Precedent: Ulrik Horn's Bioshelter Project\n\nThe closest publicly-documented precedent for a Swedish x-risk project navigating exactly this path is Ulrik Horn's **Minimum Viable Shelters for Mirror Bacteria** project, active since 2021–2022. The project's own public write-up on LessWrong — [Last Line of Defense: Minimum Viable Shelters for Mirror Bacteria](https://www.lesswrong.com/posts/2t2QuD3bPyWZ2aDBr/last-line-of-defense-minimum-viable-shelters-for-mirror) — discloses the following funding structure:\n\n- **Fiscal sponsor:** EA Sweden (the Swedish effective altruism association) provides fiscal sponsorship and administrative support.\n- **Initial funding:** Long-Term Future Fund (LTFF).\n- **Current funding:** Survival and Flourishing Fund, and Jaan Tallinn directly.\n- **Board:** Maria Khimulya and Jeff Kaufman — both long-standing, well-networked figures in the EA / AI-safety ecosystem.\n- **Timeline:** ~2 years of active work, first physical shelter unit planned for early 2025, with movement toward commercial manufacturing through a newly established company.\n\nThe topical overlap with Deliberus is zero — the shelter project is about biosecurity, not epistemic infrastructure — but the *procedural* overlap is total: Swedish base, solo-led at origin, no initial organizational scaffolding, grew into LTFF → SFF + Tallinn direct funding through EA Sweden fiscal sponsorship. Every structural question Deliberus faces (\"how do we receive a grant at all?\", \"who administers the funds?\", \"what paperwork does EA Sweden need?\", \"how did the board form?\") has a worked example in a neighboring project run by someone Deliberus can see in the same Swedish EA community.\n\nThis is documented here as a **precedent**, not an endorsement or relationship claim. The relevant insight is that the path EA-Sweden → LTFF → SFF has at least one Swedish traversal on record, which meaningfully reduces the epistemic risk of Deliberus attempting the same sequence.\n\n## 5. Long-Term Future Fund as Parallel Path\n\nThe [Long-Term Future Fund](https://funds.effectivealtruism.org/funds/far-future) (LTFF) is operated by Effective Ventures Foundation and makes grants to early-stage work that reduces long-term existential risk. Unlike SFF, which has a defined annual round, LTFF accepts applications on a rolling basis with grant decisions typically within 6–10 weeks. Grant sizes run from a few thousand dollars to several hundred thousand, with a median in the $20K–$100K range for early-stage projects. LTFF has historically been the path-finder grant for x-risk projects that later scaled into SFF, Open Philanthropy, or other larger funders — the Horn bioshelter precedent above is one example, and the pattern is common enough that it is worth treating LTFF as the first step in a standard ladder rather than as a separate backup option.\n\nFor Deliberus specifically, the LTFF framing is the AI-safety-infrastructure case already developed in [deliberus-as-alignment-infrastructure.md](deliberus-as-alignment-infrastructure.md) and [ai-safety-and-tao-augmentation-research.md](ai-safety-and-tao-augmentation-research.md).\n\nAn application pattern worth considering: LTFF first, on the strength of the alignment-infrastructure framing, to establish a track record; then SFF as the larger-scale application, referencing the LTFF grant as evidence that the work has passed one layer of x-risk-community review. The sequence is procedurally independent — you can apply to both simultaneously — but there is a narrative advantage to having LTFF in motion before SFF's recommenders evaluate the full proposal.\n\n## 6. IP and Openness Compatibility\n\nSFF's 2026 application page specifies that grantees have a default obligation to release work as \"open-source, open-access (CC-BY), and under permissive software licences (MIT + Apache 2).\" This is a requirement that most civic-tech grants only express as a preference; SFF makes it the default.\n\nDeliberus already meets this requirement structurally:\n\n- The codebase is on public GitHub at [github.com/semikolon/deliberus](https://github.com/semikolon/deliberus) and is developed in the open.\n- The documentation — including every file in `docs/` and `docs/research/` — is rendered publicly at [deliberus.com/docs](https://deliberus.com/docs) and is the same material an SFF recommender would read if they wanted to understand the project in depth before evaluating it.\n- The philosophical commitment to \"mapped perspectives, not truth\" ([vision.md §Epistemic Humility as Architecture](../vision.md)) and to transparent process (\"Focus on trusting the process instead of the people\", 2012) is not just compatible with open-access norms — it is the same commitment stated in different vocabulary.\n\nThe only place where the openness question is live is the eventual operating-entity design in Phase 2 or beyond. That design is still exploratory and the SFF application does not need to resolve it; a fiscal-sponsor path defers the question naturally. When the question becomes live, the [Mozilla Foundation + Corporation model](governance-and-financing.md) and the [Hypothes.is → Anno PBC path](governance-and-financing.md) in the original research both remain on the table, and both are compatible with open-source default releases.\n\n## 7. Risks and Caveats\n\nFour honest risks worth naming.\n\n**Single-donor concentration.** SFF is funded by Jaan Tallinn. The rounds are funder-identical across all six tracks. This is the opposite of Polis's funding fragility but has its own structural risk: a shift in the donor's priorities, health, or attention would affect the entire pipeline. This is also true of Kialo's philanthropist-patron structure, and it is a known pattern in the EA ecosystem. Mitigation: treat SFF as one input in a portfolio, not the primary source. LTFF, Open Philanthropy, Mozilla, Sloan, Hewlett, Vinnova and Horizon Europe all remain available and should be pursued on their own merits.\n\n**Grant-cycle latency.** For the live HSEE path, recommendations are announced at the end of November 2026 and disbursement is by the end of March 2027; if late routing misses 2026, the next announced S-Process round is likely SFF-2027, roughly 9-12 months later. Deliberus's self-funded runway needs to cover that gap without operational dependence on a specific outcome. A smaller LTFF, Manifund, Cosmos, EV, CAIF, or FLF grant could bridge the measurement program; absent that, SFF is still worth pursuing, but the project must be structured so that non-funding is survivable.\n\n**Speculation grant bottleneck.** The speculation-grant stage is a procedural/attention gate into the SFF ecosystem, not a near-guaranteed approval statistic. Speculators are a broader pool than the 4–12 round recommenders (~40 as of the 2026-07-30 check), but each has narrower priors and full individual discretion — most will pass on any given application quickly. The mitigation is to write the speculation submission as though it were the full proposal in compressed form — especially the 500-character pitch — not as a placeholder.\n\n**Mission drift from grant framing.** The grant application will reward the framings that SFF recommenders already believe in — AI safety, existential risk, value alignment, verification infrastructure. Those framings are genuinely compatible with Deliberus's core thesis, but they are not the *entirety* of that thesis. The cooperation synthesis ([non-zero-sum-economics-and-civilizational-cooperation.md](non-zero-sum-economics-and-civilizational-cooperation.md)), the love-through-understanding framing ([civilizational-vision.md](../civilizational-vision.md)), the attunement pole of the dialectic ([analysis-and-attunement.md](../analysis-and-attunement.md)) — these are load-bearing parts of what the project is *for*, and they should not disappear from the application in the name of grant-readability. Recommenders who would be attracted specifically to the convergence-thesis-as-alignment-substrate framing will be attracted even more strongly when that framing is named in voice the application did not need to flatten.\n\n## 8. Open Questions for the Application Itself\n\nA few questions this research doc does not settle and that the application will need to answer:\n\n- **Who is the fiscal sponsor.** The EA Sweden precedent is strong but is not a guarantee of sponsorship availability. A sponsor conversation needs to happen in the days before the application, not after.\n- **What specifically is being asked for.** SFF applications typically name a 12–24 month budget, key milestones, and a small team structure. Deliberus currently operates solo; the application needs to decide whether it is asking for solo-runway funding, a first-hire budget, or a mixed research/engineering budget, and each of those tells a different story about what the grant enables.\n- **Which framing survives the current round.** Main+HSEE stacking is rejected for the same project in 2026. The live question is how the HSEE/speculation/next-round version preserves the core thesis while staying honest about timing, routed demand, and evaluation evidence.\n- **What evidence is included.** By the time recommenders read the full proposal in the summer, the live extraction pipeline at [deliberus.com](https://deliberus.com) will be an order of magnitude more mature than it is today. An application written now but evaluated in three months needs to point to what will be visible at evaluation time, not what is visible at submission time.\n\n## 9. The Anthropic IPO Funding Landscape (Apr 17, 2026 addendum)\n\nThe SFF 2026 application window exists inside a larger capital event that makes the timing even more strategic than the round structure alone suggests.\n\n**The Anthropic IPO is imminent.** Prediction markets put the probability at ~75% within one year. Goldman Sachs, JPMorgan, and Morgan Stanley are in early discussions. The company closed a $30B Series G at $380B valuation in February 2026; secondary markets already value it at ~$850B. The expected raise exceeds $60B — second-largest technology IPO in history behind SpaceX.\n\n**The EA philanthropic capital surge that follows is historically unprecedented.** Anthropic's seven co-founders have pledged to donate 80% of their wealth. Each holds ~1.8% of the company. At the $380B Series G valuation, each pledge is worth ~$5.4B; combined co-founder pledges alone total $37.8B — nearly ten times Coefficient Giving's total historical giving. At the $850B secondary-market valuation, founder pledges could be worth $85B. Including employee pledges and other EA-aligned shareholders, total expected EA capital post-IPO is estimated at $100–200B. Three co-founders (Dario Amodei, Amanda Askell, Jack Clark) have publicly signed the Giving What We Can Pledge.\n\n**This creates strategic urgency for existing funders to deploy NOW.** Jeff Kaufman's [\"Front-Load Giving Because of Anthropic Donors?\"](https://forum.effectivealtruism.org/posts/rRBaP7YbXfZibSn3C) (EA Forum, Dec 2025) argues that near-term dollars have disproportionate impact because organizations scale more productively with smooth, front-loaded funding than with large post-IPO influxes. The mechanism: when others will donate $10M next year, donating $200K this year produces a larger marginal boost than the same amount added to the $10M pile. Sophie Kim's [\"The Anthropic IPO Is Coming. We Aren't Ready for It.\"](https://forum.effectivealtruism.org/posts/ychu2LAw54sNcocKH/the-anthropic-ipo-is-coming-we-aren-t-ready-for-it) (EA Forum, Apr 12, 2026) makes the infrastructure case: grantmaking capacity is ~30–60 evaluators globally, Coefficient Giving scales grant volume in lockstep with headcount (meaning the bottleneck is evaluator bandwidth, not project quality), and without advisory infrastructure the new donors will \"park money in donor-advised funds, where it can sit indefinitely.\"\n\n**For Jaan Tallinn specifically:** Tallinn himself is one of the largest Anthropic shareholders (he led the Series A). The IPO will make SFF's funder vastly wealthier. But the strategic calculus is the one Kaufman identifies — Tallinn's marginal dollar matters more NOW (thin funding surface, few competitors for attention) than it will post-IPO when $100B+ of new capital enters the AI-safety space. This is an argument for SFF's 2026 round being unusually well-funded and unusually receptive to strong applications.\n\n**Deliberus as a \"decorrelated bet.\"** Sophie Kim's post argues the biggest gap in the funding landscape is projects with *different threat models* than the mainline alignment labs — not another alignment research group, but infrastructure that addresses reasoning, cooperation, and epistemic integrity from a complementary angle. The post specifically calls for \"a senior person with a different threat model\" as \"the single highest-leverage addition to the ecosystem right now.\" Deliberus's pitch — structured reasoning as verification infrastructure, not alignment research itself but the epistemic substrate underneath it — IS a decorrelated bet by design. An SFF recommender who has internalized the front-loading argument and the decorrelated-bet gap will recognize Deliberus as fitting both.\n\n**The grantmaker bottleneck works in Deliberus's favor.** With 30–60 evaluators and a capacity constraint (not a quality constraint) on AI safety grants, a project that makes the evaluator's job easy — extensive public research (60+ docs at [deliberus.com/docs](https://deliberus.com/docs)), live pipeline, transparent reasoning, clear alignment-infrastructure case already articulated in [deliberus-as-alignment-infrastructure.md](deliberus-as-alignment-infrastructure.md) — has structurally better odds than the bottleneck's raw statistics suggest.\n\n## 10. Summary — Why This Matters Enough to Document\n\nThe earlier governance research correctly identified Open Philanthropy, Knight, Mozilla, Sloan, and Hewlett as possible funders, and it correctly identified Vinnova and Horizon Europe as Swedish/EU paths. It did not, however, include SFF — the one publicly-open, thesis-identical, currently-in-window grant source specifically designed to fund the kind of work Deliberus is trying to do.\n\nThat omission had a concrete cost: the recommended primary financing sequence in the earlier research put Open Philanthropy fourth in the ladder and said nothing about the April Main deadline or later HSEE timing. This research doc updates the picture: the grant ecosystem Deliberus fits inside is narrower and more time-structured than the earlier research framed it as, and the immediate action that reflects that narrower fit is to work through the fiscal sponsor, keep the speculation pitch crisp, and apply when the evidence package is strong enough rather than rushing a brittle submission.\n\n---\n\n## Sources\n\n- [SFF 2026 application page](https://survivalandflourishing.fund/2026/application)\n- [Survival and Flourishing Fund — overview](https://survivalandflourishing.fund/)\n- [Jaan Tallinn — Future of Life Institute](https://futureoflife.org/person/jaan-tallinn/)\n- [Jaan Tallinn — CSER](https://www.cser.ac.uk/team/jaan-tallinn/)\n- [Future of Life Institute — Wikipedia](https://en.wikipedia.org/wiki/Future_of_Life_Institute)\n- [Long-Term Future Fund (Effective Ventures Foundation)](https://funds.effectivealtruism.org/funds/far-future)\n- [Last Line of Defense: Minimum Viable Shelters for Mirror Bacteria — LessWrong (Ulrik Horn)](https://www.lesswrong.com/posts/2t2QuD3bPyWZ2aDBr/last-line-of-defense-minimum-viable-shelters-for-mirror)\n- [Ulrik Horn's Quick Takes — EA Forum](https://forum.effectivealtruism.org/posts/DynkQp3CZCpAyzsQS/ulrik-horn-s-quick-takes)\n- [80,000 Hours AI-enhanced decision-making problem profile](80000-hours-ai-decision-making.md) *(in-repo; summarizes the 80K Hours source)*\n- [Deliberus as Alignment Infrastructure](deliberus-as-alignment-infrastructure.md) *(in-repo)*\n- [Governance and Financing for Deliberus](governance-and-financing.md) *(in-repo; broader org-models context, pre-SFF-finding)*\n\n*Sources added in the Apr 17, 2026 addendum:*\n- [\"The Anthropic IPO Is Coming. We Aren't Ready for It.\" — EA Forum (Sophie Kim, Apr 12, 2026)](https://forum.effectivealtruism.org/posts/ychu2LAw54sNcocKH/the-anthropic-ipo-is-coming-we-aren-t-ready-for-it)\n- [\"Front-Load Giving Because of Anthropic Donors?\" — EA Forum (Jeff Kaufman, Dec 3, 2025)](https://forum.effectivealtruism.org/posts/rRBaP7YbXfZibSn3C)\n- [\"Anthropic employees say they'll give away billions\" — Transformer News](https://www.transformernews.ai/p/anthropic-employees-philanthropy-billions-donations-effective-altruism-coefficient-giving-ai-safety)\n- [Anthropic IPO 2026 — EBC Financial Group](https://www.ebc.com/forex/anthropic-ipo-2026-latest-timeline-valuation-and-risks)\n- [Anthropic's rise is giving some OpenAI investors second thoughts — TechCrunch (Apr 14, 2026)](https://techcrunch.com/2026/04/14/anthropics-rise-is-giving-some-openai-investors-second-thoughts/)\n\n*Sources added in the Jul 9, 2026 S-Process / Manifund / Surplus update:*\n- [SFF-2024 S-Process Recommendations Announcement](https://survivalandflourishing.fund/2024/recommendations)\n- [Podcast: Austin and Oli on funding & incubating projects — EA Forum linkpost](https://forum.effectivealtruism.org/posts/mFoK2kjauoGxsEFaQ/podcast-austin-and-oli-on-funding-and-incubating-projects)\n- [Surplus: for massive public good — Austin Chen](https://manifund.substack.com/p/surplus-for-massive-public-good)\n- [Manifund regranting FAQ](https://manifund.org/about/regranting)\n- [Manifund donation FAQ](https://manifund.org/about/donate)\n"}