Effective Altruism / SBF / OpenAI-Safety Funding Network (2014–23)

Introduction

Between roughly 2014 and 2023, a dense web of funding, personnel, and ideology connected the Effective Altruism (EA) movement, AI-safety research organisations, and the two most prominent commercial AI labs in the world. Open Philanthropy — the philanthropic vehicle of Dustin Moskovitz and Cari Tuna — provided hundreds of millions of dollars to organizations including OpenAI, Anthropic, MIRI (Machine Intelligence Research Institute), and the Centre for Effective Altruism. FTX founder Sam Bankman-Fried (SBF) publicly identified as an EA adherent and in 2022 committed the FTX Future Fund to over $1 billion in grants, the bulk aimed at AI-safety and EA-adjacent causes, before FTX collapsed in November 2022 and the Future Fund dissolved.

The conspiracy framing asks whether this network amounted to coordinated ideological capture of AI governance — a small, like-minded, Bay Area rationalist community positioning itself as the gatekeeper of civilisation-scale decisions about artificial intelligence.

The Network: What Is Documented

Open Philanthropy is the most important institutional node. Founded in 2012 by Moskovitz (Facebook co-founder) and managed by Holden Karnofsky and then Alexander Berger, it made grants exceeding $100 million to OpenAI in 2017 and funded Anthropic, MIRI, the Future of Humanity Institute (FHI) at Oxford (founded by Nick Bostrom), and numerous EA movement-building organisations. The personnel overlap is substantial: Karnofsky married Dario Amodei's sister; multiple OpenAI and Anthropic employees held 80,000 Hours career connections.

The Centre for Effective Altruism (CEA), 80,000 Hours, and the Global Priorities Institute (Oxford) form the EA infrastructure. FHI under Bostrom published the academic work (including Superintelligence, 2014) that provided the intellectual foundation for existential-risk-focused AI-safety work — the same framing that shapes OpenAI's and Anthropic's stated missions.

FTX Future Fund and SBF

In 2022, SBF publicly announced that FTX and the FTX Future Fund would commit over $1 billion to EA and AI-safety causes. The Future Fund made tens of millions in grants before FTX's November 2022 collapse. SBF was simultaneously making large political donations — reportedly over $100 million across the 2022 cycle — which some critics characterised as politically motivated reputation-laundering to forestall regulatory scrutiny of crypto. He was convicted of fraud and related charges in November 2023.

The collapse of FTX caused significant disruption to EA-funded organisations that had received or were expecting Future Fund grants. It also generated intense scrutiny of the EA movement's ethics and internal culture.

Helen Toner and the OpenAI Board

In November 2023, OpenAI CEO Sam Altman was briefly fired by the OpenAI board before being reinstated days later following a staff revolt. Helen Toner, a board member with Georgetown CSET (Center for Security and Emerging Technology) and Open Philanthropy connections, was identified in subsequent reporting as a driver of the board action. A paper she co-authored had criticised OpenAI's safety communications. The episode illustrated real tensions between EA-aligned board members and commercial leadership — and led to a board reconstitution that reduced EA influence. Gideon Lewis-Kraus profiled the EA-AI nexus for The New Yorker in March 2023, providing the most detailed journalistic account of the network.

The Conspiracy Framing

The claim is not simply that an EA funding network existed — it did, and it is well-documented. The conspiracy framing holds that this constituted:

• Coordinated ideological capture: a small community intentionally placing its members in positions of influence over AI governance to advance a specific worldview (long-termism, existential risk prioritisation) that privileges far-future speculative harms over present harms.
• Regulatory arbitrage: using AI-safety framing and political donations to shape regulation in ways that benefit incumbents and freeze out competition.
• SBF as conduit: the claim that FTX donations to politicians were not altruistic but were designed to purchase regulatory forbearance for crypto while FTX operated fraudulently.

What the Evidence Shows

The network existed. The funding flows are documented in 990 filings, grant databases, and journalism. The personnel overlaps are real. The intellectual influence of EA/longtermism on OpenAI's and Anthropic's stated missions is real and self-acknowledged. The SBF fraud is a matter of criminal conviction.

What is disputed is whether the network constituted intentional coordinated capture rather than an organic clustering of like-minded people around shared ideas in a small geographic and intellectual community. The Bay Area rationalist community is genuinely small; the overlap between EA, AI-safety, and the leading AI labs reflects shared intellectual origins more than active conspiracy.

Verdict

Confirmed as a real, well-documented funding and influence network. The framing of intentional ideological "capture" is contested; the network's existence, scope, and effects on AI governance are matters of documented public record, not speculation.

What Would Modify Our Assessment

• Internal communications showing explicit coordination to exclude competing AI-safety frameworks
• Evidence that Open Philanthropy funding came with ideological conditions attached to grantees' governance positions
• Documentary evidence that SBF political donations were coordinated with EA leadership as a regulatory strategy

FTX's Direct Stake in Anthropic: A Documented Financial Link

The existing record covers Open Philanthropy's funding of OpenAI and MIRI, but the single most concrete financial link between SBF/FTX and a frontier AI lab is FTX's direct equity investment in Anthropic, which is often underreported relative to the EA-adjacent grant flows. In April 2022, FTX put $500 million into Anthropic — then a nascent AI-safety-focused startup founded by ex-OpenAI researchers Dario and Daniela Amodei — acquiring roughly an 8% stake as part of the company's Series B round. This was not a philanthropic grant; it was a for-profit venture investment made directly by the exchange that Bankman-Fried controlled, entangling a company under criminal fraud investigation with one of the two dominant AI-safety labs as a shareholder of record.

When FTX collapsed in November 2022, the Anthropic stake became an asset of the bankruptcy estate. In February 2024, administrators sold it in stages for a combined $1.3 billion — a roughly 2.6x return that went toward repaying defrauded FTX customers, not toward funding further EA or AI-safety work. This detail cuts against a simple "SBF used crime proceeds to buy influence over AI" narrative in two ways: first, the sale returned the money to fraud victims rather than to any EA cause; second, in hindsight the estate significantly undersold the position — the same 8% stake would be worth on the order of $30 billion at Anthropic's later valuations, meaning the bankruptcy trustees prioritized speed of creditor repayment over maximizing the value that critics might otherwise point to as evidence of a lucrative "AI-safety cartel" payoff. The mundane explanation (a venture bet that appreciated, managed by court-appointed trustees with fiduciary duties to defrauded customers) fits the facts better than a coordination narrative.

Karnofsky's Recusal: Evidence Against, Not For, Coordinated Capture

The personnel overlaps between Open Philanthropy and the AI labs are real, but the documented governance history shows friction and self-imposed limits, not a smooth capture operation. Holden Karnofsky, Open Philanthropy's co-founder, took an OpenAI board seat in 2017 as part of the $30 million grant relationship. When his wife, Daniela Amodei, left OpenAI to co-found Anthropic — a direct OpenAI competitor — in 2021, Karnofsky resigned his OpenAI board seat, citing the conflict of interest. Open Philanthropy then nominated Helen Toner, a Georgetown CSET researcher with her own Open Philanthropy history, to take the seat. This is a sequence of disclosed conflicts and formal recusals, which is the opposite of what a covert-influence operation would look like; a genuinely coordinated capture effort would not require a principal to publicly step down over an obvious conflict. It does, however, confirm how small and tightly networked the community of people making these decisions really is — which is itself worth stating plainly rather than either minimizing or dramatizing.

Coefficient Giving: The Rebrand and Continuity Question

In November 2025, Open Philanthropy renamed itself Coefficient Giving. According to the organization's own announcement and reporting on it, the impetus was mundane and administrative: staff reported that journalists and prospective grantees confused "Open Philanthropy" with OpenAI itself, or with the separate Open Society Foundations, and the organization had grown beyond being the exclusive giving vehicle of Dustin Moskovitz and Cari Tuna's Good Ventures, now channeling more than $100 million a year from other donors. The organization states that its AI-safety-focused "Navigating Transformative AI" program, staff, and grantmaking criteria carried over unchanged under the new name.

The rebrand matters for this theory in two ways. First, it is a reminder that names change while institutional relationships and funding commitments often do not — researchers checking whether "Open Philanthropy still funds OpenAI" a year from now need to know to look for Coefficient Giving instead, or they will wrongly conclude the relationship ended. Second, some commentary has framed the renaming itself as evidence that the organization sought to obscure its ties following the reputational damage of the SBF collapse and the OpenAI board crisis; the organization's own stated rationale (brand confusion, expanding beyond one anchor donor) is a more mundane and better-supported explanation, though the timing — two years after the OpenAI crisis and three years after FTX's collapse — invites the inference regardless. Both are worth recording: the confirmed rationale as stated by the organization, and the surrounding skepticism, without asserting the skeptical reading as fact.

SBF's Appeal Exhausted: What the Legal Record Actually Settles

On appeal, Bankman-Fried argued the trial court wrongly excluded evidence that his investments were fundamentally sound and that customers would eventually have been repaid had FTX not collapsed when it did. In June 2026, a three-judge panel of the Second Circuit Court of Appeals rejected this argument and upheld his conviction on all counts, writing that fraud occurs when a defendant uses a material misstatement to induce a victim to hand over money, regardless of whether the defendant intended for the victim to ultimately lose it. This closes the direct appellate path (further review by the full Second Circuit or the Supreme Court remains theoretically available but is discretionary and rare).

This matters for keeping this theory precise: the appellate ruling confirms the fraud conviction and forecloses SBF's own "it would have worked out" defense, but it says nothing about the separate, contested question of whether EA or AI-safety organizations knowingly participated in or benefited from fraudulent conduct. No EA-affiliated leader, Open Philanthropy/Coefficient Giving staffer, or AI-lab executive has been charged with a crime in connection with SBF's fraud. The legal record settles SBF's own culpability; it does not settle — because it was never asked to settle — the interpretive question of institutional complicity that the broader theory raises.

EA Leadership's Own Reckoning

It is documented that senior EA figures received specific warnings about Bankman-Fried's conduct years before FTX's collapse. Reporting found that as early as 2018, Alameda Research colleagues confronted Bankman-Fried over what they described as gross negligence and dishonesty, leading roughly half of the firm's staff to resign; people close to EA leadership, including co-founder William MacAskill, moral philosopher Nick Beckstead, and Open Philanthropy's Holden Karnofsky, were told about these concerns in the years that followed, according to named sources. In the nine months before FTX's collapse in 2022, the FTX Future Fund had distributed more than $160 million to EA-affiliated causes, including over $33 million to organizations connected to MacAskill, such as the Centre for Effective Altruism and Longview Philanthropy.

After the collapse, MacAskill publicly called his friendship with and promotion of Bankman-Fried "a matter of great shame," stated that Bankman-Fried "did an enormous amount of harm," and argued that a clear-thinking EA framework should have "strongly opposed" the ends-justify-the-means reasoning SBF is alleged to have used to rationalize fraud. This combination — documented advance warnings that were reportedly downplayed, followed by public post-collapse contrition — is itself relevant evidence: it undercuts the idea that EA leadership was a knowing co-conspirator (they distanced themselves and expressed genuine reported shock at the extent of the fraud) while also undercutting the idea that the ties were merely incidental (specific, credible warnings existed and were not acted on before large sums changed hands).

Distinguishing Documented Funding From Speculative Control

Across all of this, the clearest analytical line to draw is between two separate claims that are frequently blurred in conspiratorial retellings:

1. Documented claim: EA-aligned philanthropy (Open Philanthropy/Coefficient Giving) and EA-adjacent capital (FTX, SBF personally) provided substantial, traceable funding to OpenAI, Anthropic, MIRI, and EA infrastructure organizations, and EA-aligned individuals held board seats and leadership roles at these organizations at various points. This is supported by grant records, SEC/bankruptcy filings, and the organizations' own public statements.
2. Speculative claim: This funding network was assembled or used as a deliberate mechanism of ideological control over the development and regulation of AI, in a coordinated way that individual actors within it understood and executed together.

The first claim is well-supported by primary records. The second requires evidence of intent and coordination that has not been publicly documented — the available record instead shows organic overlap (a small, geographically and intellectually concentrated community), disclosed conflicts of interest that were sometimes resolved through recusal (Karnofsky), and internal disputes and warnings that were reportedly ignored rather than orchestrated (the pre-FTX-collapse warnings about SBF). Readers should treat any claim of coordinated intent as the contested part of this theory, separate from the funding relationships themselves, which are not in serious dispute.