Longview Philanthropy

Stated Theory of Change

Longview's stated theory of change: AI risk, nuclear risk, biosecurity risk, and other global catastrophic risks are underfunded relative to their severity. The bottleneck is not the supply of good organizations to fund but the supply of informed, engaged donors. Longview exists to bridge this gap by providing free, expert-led philanthropic advisory to major donors ($1M+/year, $100K+ for AI) and by managing pooled funds that allow donors to delegate decisions to Longview's grantmakers.

The causal chain: (1) Recruit and educate UHNW donors on catastrophic risks; (2) Investigate the highest-impact funding opportunities using a rigorous two-stage process with explicit "best case against"; (3) Recommend grants to donors or disburse through pooled funds; (4) Monitor grantee performance; (5) Create new projects and organizations where gaps exist.

This is a meta-level theory of change -- Longview does not directly reduce AI risk; it multiplies the impact of philanthropic capital by directing it more effectively. The claim is that Longview's marginal dollar of operational spending unlocks many dollars of well-directed philanthropy.

Revealed Theory of Change

The actions broadly align with the stated theory, with some important nuances:

AI is the dominant priority. Despite listing AI, nuclear, biosecurity, and digital sentience as focus areas, AI clearly dominates. The $133M+ directed toward AI risk reduction, the FAIF ($20M raised), the HEMs RFP ($2-10M), and the hiring of 6 AI grantmakers vs. 4 nuclear grantmakers all point to AI as the primary focus. This is consistent with EA priority-setting and donor demand.

Private funds are growing relative to public. The trajectory from the public Longtermism Fund (renamed ECF) to the private FAIF suggests Longview is shifting toward higher-dollar, lower-transparency vehicles. This may reflect genuine strategic reasoning (private funds can move faster, support sensitive projects, and serve UHNW donors better) but it also reduces public accountability.

Donor cultivation is the core activity. The CPO job listing, the Napa and Stockholm summits, and the emphasis on "several dozen UHNW donors who collectively have the potential to give hundreds of millions" reveal that donor relationship management is as important as grantmaking research. Longview is as much a donor services organization as it is a research-driven funder.

Project creation is a differentiator. Longview does not only recommend existing grantees -- it creates new projects (SecureDNA sponsorship, OECD partnership, Digital Sentience Consortium, HEMs research agenda). This means Longview shapes what exists to be funded, not just which existing orgs get funded. This is more influential and more opaque than simple grantmaking.

CG dependence is structural. $22.25M in CG operational support (72% of the single largest grant alone) means Longview's existence as an organization depends on CG's continued support. The stated goal of being an "independent alternative" to CG is in tension with this funding dependency.

Key Assumptions

Assumption 1: UHNW donors can be recruited and retained at scale.

• Evidence for: $133M+ directed, 3 Giving Pledge connections, FAIF raised $20M in 9 months, 2000+ ECF donors.
• Evidence against: No published donor retention data. Anticipated Anthropic windfall is uncertain. Growth from 24 to 35+ staff requires proportional donor growth.
• Testable? Yes -- track annual capital directed.
• If wrong: Longview becomes a small funder with high overhead, losing the leverage that makes the advisory model work.

Assumption 2: Longview's grantmaking is genuinely better than what donors would do on their own.

• Evidence for: Two-stage investigation process, "best case against" each grant, expert program officers (Robichaud, O'Gara), CG endorsement, GWWC recommendation.
• Evidence against: No external evaluation of grantmaking quality. No published grant outcomes data. No comparison to alternative advisory services.
• Testable? In principle, by comparing outcomes of Longview-advised grants to a counterfactual portfolio. In practice, very difficult.
• If wrong: Longview adds overhead without improving capital allocation.

Assumption 3: AI safety funding diversification reduces ecosystem risk.

• Evidence for: CG explicitly says the field needs more independent funders to "correct our blind spots." Monoculture funding creates single points of failure and worldview lock-in.
• Evidence against: If Longview shares deal flow with CG and draws from the same donor pool (EA-aligned tech wealth), the "diversification" may be more nominal than real. PauseAI exclusion criticism suggests the Overton window of fundable approaches may be the same regardless of which intermediary routes the money.
• Testable? Compare Longview's grantee portfolio to CG's. If they overlap >70%, diversification value is low.
• If wrong: Longview adds complexity without genuine strategic independence.

Assumption 4: The advisory model scales efficiently.

• Evidence for: Advising costs are modest relative to capital directed ($3.3M US expenses vs. $133M+ directed).
• Evidence against: Growth requires expensive senior hires (CPO, AI grantmakers). UHNW donor relationships are labor-intensive and require high-touch service.
• Testable? Track operational cost ratio over time.
• If wrong: Longview becomes a boutique advisory with high costs per dollar directed.

Strengths

Genuine niche. The gap between GiveWell (small donors, static list) and full foundations (years to build) is real. Longview fills it with a model that provides research-quality grantmaking recommendations for free to donors who don't have the time or expertise to find opportunities themselves.

CG ecosystem endorsement. CG's public acknowledgment that Longview provides needed independent capacity, combined with CG referring donors to Longview and explicitly sharing deal flow, gives Longview an unusual institutional validation.

Technical depth in specific domains. The HEMs RFP demonstrates genuine technical expertise in hardware governance for AI. Carl Robichaud's decade at Carnegie brings real nuclear policy depth. These are not generic philanthropic advisors.

Project creation capacity. The ability to create new projects (SecureDNA, OECD partnership, Digital Sentience Consortium, HEMs agenda) rather than just recommend existing ones gives Longview outsized influence relative to the capital it directly controls.

Rapid growth and ambition. Moving from a small EVF project to an independent organization directing $133M+ in 8 years, with a $100M+/year target, demonstrates execution capacity.

Weaknesses and Risks

CG funding dependency. The central irony: an organization whose value proposition is "independent alternative to CG" is primarily funded by CG. If CG shifted priorities or reduced support, Longview's operations would be at risk. The $15.96M grant in October 2024 represents years of operational runway, but the dependency is structural.

Transparency deficit. No published COI policy. No public FAIF grantee list. No annual report. No grant concentration data. No external evaluation. For an organization directing $100M+/year, this is a significant governance gap. The private advisory model inherently limits transparency, but the private pooled funds (which Longview directly controls) could be more transparent.

Funder capture risk. If Longview's donors are AI company employees (as anticipated with Anthropic wealth), and Longview recommends grants to organizations that evaluate those AI companies, a circular funding dynamic emerges. Longview has no published policy for managing this conflict.

Limited public accountability for advisory recommendations. Most of Longview's $133M+ in directed capital goes through private advisory channels. Donors can accept or reject recommendations. There is no public accountability for what was recommended, what was accepted, or what outcomes resulted.

Ecosystem groupthink. Longview's staff come heavily from the GovAI/Oxford pipeline. Donors are EA-aligned. CG refers deal flow. MacAskill and Karnofsky have endorsed the organization. This tight network could produce excellent grantmaking if the worldview is correct, or systematically blind spots if the worldview has gaps. The absence of any evidence that Longview funds confrontational AI safety approaches is a specific concern.

Ben Delo reputational risk. Longview's most prominent named donor pled guilty to a Bank Secrecy Act violation (failing to implement AML at BitMEX), paid a $10M penalty, and was pardoned by Trump. While the legal matter is resolved, the association could be used against Longview by critics of crypto wealth in philanthropy.

Cross-References

Coefficient Giving: Primary operational funder and deal-flow sharing partner. CG holds the dominant share of AI safety funding; Longview is the primary "independent alternative" that CG itself supports. The relationship is cooperative rather than competitive.

AISTOF (AI Safety Tactical Opportunities Fund): Peer donor advisory fund that CG also names as an independent alternative. AISTOF reportedly filled gaps before CG scaled up technical safety funding.

METR: Known Longview grantee (via old Longtermism Fund, $220K to then-ARC Evals). METR has a published policy of not accepting money from AI companies or their executives -- a standard Longview lacks.

PauseAI: Not a known Longview grantee. Holly Elmore's criticism of CG's non-funding of PauseAI extends to the broader ecosystem Longview operates in. Whether Longview would fund confrontational approaches is the key test of genuine independence.

GovAI (Centre for the Governance of AI): Multiple Longview AI staff (O'Gara, Mehta) come from GovAI. This Oxford pipeline shapes Longview's governance-oriented approach to AI safety.

Giving What We Can: ECF partnership. GWWC has evaluated and recommended ECF but not Longview's broader operations.

Founders Pledge: Funds Longview. Operates in the same space of advising tech founders on effective giving.

What Would Change This Assessment

Strong evidence of independence from CG: If Longview published a grantee portfolio showing significant non-overlap with CG's portfolio -- especially funding organizations CG has declined -- this would substantially validate the "independent alternative" claim.

Published COI policy and annual report: If Longview published a conflict of interest policy, an annual report with consolidated financials, and aggregate grantee data (even without individual names), this would address the transparency concerns.

Evidence of funding confrontational approaches: If Longview were shown to fund organizations like PauseAI, Evitable, or other groups that challenge AI lab practices (rather than supporting them), this would demonstrate that the donor advisory model does not systematically exclude uncomfortable recommendations.

CG funding phase-out: If Longview reached a point where CG operational support was <20% of total revenue, the independence claim would be substantially more credible.

Major grant failure or controversy: If a Longview-recommended grant were publicly revealed to have failed badly or created harm, this would test the organization's governance and learning capacity.

UK financial data: Access to Longview Inc. Ltd's Companies House filings would complete the financial picture and reveal the full scale of operations.

Self-Critique

Sources I should have checked but didn't:

• Longview Inc. Ltd's Companies House filings (UK entity financials)
• The EA Forum post "Longview is hiring: What Longview is like (from my perspective) in 2025" (blocked domain, only seen in snippets)
• The full 2024 and 2025 ECF/FAIF annual reports (PDFs not fully extractable)
• Whether any Longview board members have financial interests in AI safety organizations

Where this analysis is potentially biased:

• I may overweight the CG dependency concern. CG funding Longview could be straightforwardly good: the best-informed AI safety funder investing in building alternatives to itself is exactly what you'd hope for.
• I may underweight Longview's actual grantmaking quality because of limited visibility into private channels. The HEMs RFP suggests genuine sophistication, and the known grantees (METR, MATS, CHAI, OECD) are credible organizations.

What a thoughtful person who disagrees would say: "You're applying for-profit governance standards to a small nonprofit. Longview is 8 years old with 24 people. The governance gaps you identify (no COI policy, no annual report) are normal for organizations at this stage. The CG relationship is not dependence -- it's the field's most informed funder making a rational investment in diversification. And the opacity of the advisory model is a feature, not a bug: donors need confidentiality to give freely."

My single weakest claim: That the CG-Longview deal flow sharing arrangement compromises Longview's independence. I have no evidence of how this works in practice. It's possible that CG shares opportunities and Longview genuinely evaluates them independently, funding some that CG wouldn't. Without seeing the internal process, this concern is speculative.

What information would most change my view: Seeing Longview's full grantee portfolio (across advisory and pooled funds) compared to CG's portfolio. If the overlap is low and includes organizations CG has declined to fund, my assessment of Longview's value would increase substantially. If the overlap is high, the "independent alternative" claim is hollow.