Effective Ventures Foundation

Stated Theory of Change

EV existed to reduce the friction of launching and operating EA-aligned organizations. By providing fiscal sponsorship, shared operations, legal structure, and governance, EV would enable the organizations that directly reduce catastrophic risks -- AI labs, safety researchers, policy groups, evaluators -- to focus on their core missions. EV's parent entity, CEA, had the further mission of growing and maintaining the EA community itself: running conferences for talent matching, maintaining the EA Forum for epistemic discourse, supporting local groups, administering EA Funds for grantmaking, and overseeing community health.

The causal chain: CEA grows the community -> community produces talented people -> talented people work at safety orgs -> safety orgs reduce AI risk. Additionally: EA Funds distributes money -> safety orgs get funded -> safety orgs do safety work. EV was the organizational wrapper that made all of this legally and operationally possible.

Open Philanthropy funded this theory to the tune of $177M across 43 grants, classifying nearly all of it under "Global Catastrophic Risks Capacity Building."

Revealed Theory of Change

The actions reveal a theory of change that was less about growing community impact and more about building physical and social infrastructure for an in-group:

Infrastructure over impact. Approximately $55M of OP grants (~31% of total) went to real estate: Wytham Abbey ($23.2M), Harvard Square coworking ($24M), Oxford office space ($2.65M), biosecurity coworking ($5.3M). This is an extraordinary allocation for a community-building organization. Wytham Abbey, purchased as a "conference venue modeled on the Rockefeller Foundation's Bellagio Center," was sold at an $11-14.5M loss. The revealed preference was for creating high-end physical spaces for EA elites rather than maximizing community reach or direct impact.

Governance as afterthought. Revenue grew 9x (from $8.4M to $75.8M) between 2019 and 2022. During this period, EV had no CEOs, one in-house lawyer (hired months before FTX collapsed), no finance team, no formal COI disclosure for trustees, and financial policies that were "not adhered to or reviewed regularly." The revealed theory was that smart, values-aligned people would figure it out without institutional infrastructure -- a startup mentality applied to managing tens of millions in charitable funds.

Donor capture. EA leaders were warned about SBF in 2018-2019 and dismissed the warnings. MacAskill vouched for SBF to Elon Musk in 2022. Nick Beckstead, an EV trustee, simultaneously served as CEO of FTX Foundation. The revealed theory was that access to major donors was more important than accountability -- that the money SBF could provide outweighed the ethical red flags reported by multiple people.

Community health as liability management. Julia Wise's community health work, as described in her own public statements and the TIME investigation, focused on "weighing the possible harm to the accused" against "the possible harm to other people." The 2026 Fran case shows this approach persisting: harassment was documented, investigated, and confirmed, but the institution prioritized minimizing disruption over protecting the victim. The revealed theory was that maintaining social cohesion within the community mattered more than addressing misconduct decisively.

CEA's 2025-26 "stewardship" strategy represents a genuine pivot. It explicitly names growth, brand improvement, and funding diversification as priorities. But it retains the core assumption that CEA should be the movement's steward without any democratic mandate from the community.

Key Assumptions

Assumption 1: EA community building is the best use of these resources for reducing AI risk.

• Evidence for: EA has produced many of the people now working in AI safety. 80,000 Hours careers advice was influential. GWWC has moved $253M+ to effective causes.
• Evidence against: No outcome evaluation has ever been conducted. CEA cannot demonstrate how many AI safety researchers its programs produced. The $55M in infrastructure spending could have directly funded dozens of safety research groups. The theory of change is unfalsifiable by design.
• Is it testable? Yes, in principle. Track career outcomes of EAG attendees, EA group alumni, etc. CEA has not done this.
• What changes if wrong? If community building is not cost-effective, then the $177M from OP was a massive misallocation that could have funded direct AI safety work.

Assumption 2: A single steward organization can effectively serve a diverse community.

• Evidence for: CEA has the infrastructure (Forum, conferences, Funds) and institutional knowledge.
• Evidence against: The fiscal sponsorship model correlated all risks. The Owen case, FTX, and Wytham all metastasized because they were in the same entity. The "principles-first" stance conflicts with the reality that most funding is for AI/longtermism.
• What changes if wrong? The EA community may be better served by multiple competing steward organizations rather than a monopoly.

Assumption 3: Post-FTX reforms were sufficient.

• Evidence for: 100% board turnover, formal COI policies, finance team, donor due diligence, whistleblowing policies.
• Evidence against: The 2026 Fran case occurred after all reforms. CEA "failed to take reasonable steps to prevent sexual harassment" -- language from a formal legal finding. Policy updates without cultural change are insufficient.
• What changes if wrong? CEA's claim to be a reformed organization loses credibility, and the community may need external accountability mechanisms.

Assumption 4: Open Philanthropy's dominance is manageable.

• Evidence for: OP itself acknowledges the risk and is trying to reduce dependency through matching programs.
• Evidence against: The OP-EV revolving door (CEO, board members, staff) makes genuine independence structurally impossible. OP funded EV's growth without demanding governance capacity.
• What changes if wrong? The EA movement's institutional infrastructure is effectively an extension of Open Philanthropy's will, not an independent community.

Strengths

Honest self-assessment. Robinson's reflections post is remarkably candid by institutional standards. CEA's "Our Mistakes" page exists (though critics note it understates problems). The willingness to publish post-mortems sets a higher standard than most nonprofits or corporations.

The spin-out decision was correct and courageous. Winding down EV and decentralizing the ecosystem was the right call. It decorrelates risks, gives projects their own governance, and forces each org to build its own institutional capacity.

Concrete outputs. GWWC's $253M in facilitated donations is real, measurable impact. EA Global conferences are the primary talent-matching venue for AI safety. The EA Forum is the primary public discourse platform for the field. EA Funds distributes $30M+/year to safety organizations.

Growth recovery. 25% growth in engagement across all tiers in 2025 suggests the community is resilient. EA Global Bay Area 2025 was the biggest US event since 2022.

New leadership quality. Robinson, whatever his OP ties, has demonstrated operational competence and intellectual honesty that previous leadership lacked.

Weaknesses and Risks

Repeated harassment failures. Three separate instances: the TIME-reported cases (2023), Owen Cotton-Barratt (2021 report, 2023 board awareness), and Fran's case (2024-2026). Each occurred in a different time period. Each prompted policy reforms. Each successor case demonstrated the reforms were insufficient. This is the clearest evidence of a cultural problem rather than a policy problem.

No accountability to the community. CEA claims stewardship but has no democratic mandate. The board is self-selecting. The community has no mechanism to constrain, redirect, or remove CEA. The structural critique -- that CEA holds power without accountability -- has not been addressed.

Funding monoculture. Near-total dependence on Open Philanthropy is existential. Despite acknowledging this, CEA's diversification efforts are nascent. The matching program is a start but does not address the structural revolving door between OP and CEA leadership.

Infrastructure waste. The $55M in real estate spending, including a $14.5M loss on Wytham Abbey, represents an enormous misallocation of charitable funds. This money could have funded dozens of AI safety research groups for years.

Unfalsifiable theory of change. CEA has never demonstrated that its programs produce measurable AI safety outcomes. It reports engagement metrics (conference attendees, Forum users, group members) but not impact metrics (career changes, research produced, policy influenced). For a movement built on evidence-based decision-making, this is a fundamental credibility gap.

Institutional silence from key figures. MacAskill has not published a detailed written account of his role in the SBF relationship. No community-governed investigation has occurred. The people who made governance mistakes are the ones deciding whether to investigate. This undermines trust even when the institutional reforms are genuine.

Cross-References

Open Philanthropy: EV/CEA's primary funder ($177M) and the source of multiple board members and staff. The relationship is symbiotic but creates structural dependency and conflicts. OP bears significant responsibility for enabling EV's governance-capacity gap.

80,000 Hours: The most successful direct-impact project that was housed in EV. Now independent. Its career advisory work directly connects to AI safety talent pipelines. The spin-out is healthy.

Giving What We Can: The most measurably impactful EV project. $253M in facilitated donations. Spinning out. Its pledge model is a genuine innovation in philanthropic infrastructure.

EA Funds: Merging into CEA, creating a combined community builder + funder. This mirrors the problematic dynamics of the old EV structure, concentrating community and funding power in one entity. Comparison to Longview Philanthropy (independent grantmaker) is instructive.

FHI: Housed in the EV ecosystem but closed due to Oxford University bureaucracy, not EV failures. Its closure is a loss for the field unrelated to EV's governance problems.

All AI safety orgs funded through EA Funds: The merge means CEA will control both community programming AND primary funding for many safety organizations. This concentration deserves scrutiny.

What Would Change This Assessment

• Evidence that CEA programs produce measurable AI safety outcomes. If CEA published a rigorous career-tracking study showing X% of EAG attendees entered safety work, that would substantially strengthen the theory of change.
• Successful funding diversification. If CEA reduced OP dependency below 50% while maintaining program quality, that would address the monoculture risk.
• A community governance mechanism. If CEA created an elected advisory board, binding community votes on strategic direction, or other democratic accountability mechanisms, that would address the structural critique.
• Clean handling of the next misconduct case. If a future allegation is handled promptly, fairly, and without requiring public pressure, that would demonstrate cultural change beyond policy change.
• MacAskill publishing a detailed written account. If the movement's founder gave a specific, falsifiable account of what happened with SBF and what he learned, that would rebuild trust.
• Evidence that the EA Funds/CEA merger does NOT reproduce EV's concentration of power. If the merged entity has strong internal firewalls and independent governance for grantmaking decisions, that would mitigate the structural risk.

Self-Critique

Limitations of this analysis:

• UK financial data is incomplete. The analysis relies heavily on US 990 data. EV UK's detailed financials would provide the other half of the picture.
• No insider interviews. Everything here is from public sources. Conversations with current CEA staff, spin-out project leaders, or former board members would add crucial context.
• Potential negativity bias. The evidence base is weighted toward crises (FTX, Owen, Wytham, Fran) because crises generate public documentation. CEA's quiet, effective work -- connecting someone to a safety role at a conference, funding a promising researcher through EA Funds -- is invisible in public sources.
• The counterfactual is genuinely hard. If EV had never existed, would the EA movement have developed differently? Would the organizations have launched independently, perhaps with better governance? Or would the movement have failed to coalesce? This analysis assumes EV was flawed but does not assess whether it was net positive.

What a thoughtful defender would say: "You're judging a decade-old institution by post-crisis standards. EV was a startup that grew fast in a nascent movement. Yes, governance lagged growth -- that happens at every fast-growing org. The key question is what happened after the crisis, and the answer is: comprehensive reform, honest post-mortems, and structural decentralization. Name another nonprofit that achieved 100% board turnover, published detailed reflections on its failures, and voluntarily dissolved itself. EV made mistakes but its response was admirable."

This is a fair point. Robinson's post-mortem is genuinely impressive by institutional standards. The challenge is that the pattern of requiring external pressure before acting (TIME articles, Charity Commission inquiry, Fran's public post) undermines the narrative of proactive reform. And the 2026 Fran case, occurring after all reforms, suggests the cultural problems are deeper than any policy can address.

Single weakest claim: My assessment that CEA's theory of change is "unfalsifiable." CEA could, in principle, commission an outcome study. The absence of such a study does not prove the theory is wrong -- it proves it is untested. The distinction matters.