OpenAI Foundation Governance Paradox
Quick Assessment
Section titled “Quick Assessment”| Aspect | Assessment |
|---|---|
| Governance Structure | Nonprofit holds Class N shares with exclusive board appointment power |
| Same-Board Problem | 7 of 8 Foundation directors also serve on for-profit board |
| External Capture Protection | Strong—Class N shares cannot be purchased |
| Internal Capture Protection | None—same people control both entities |
| Board Incentive Alignment | Strongly favor stock appreciation over charitable deployment |
| Post-IPO Implications | Public shareholders get economic rights, zero governance power |
The Core Paradox
Section titled “The Core Paradox”The OpenAI FoundationOpenai FoundationThe OpenAI Foundation holds 26% equity (~\$130B) in OpenAI Group PBC with governance control, but detailed analysis of board member incentives reveals strong bias toward capital preservation over p...Quality: 87/100 holds Class N Common Stock giving it exclusive power to appoint and remove all board members of OpenAI Group PBC. On paper, this means a nonprofit dedicated to “beneficial AI for humanity” controls one of the world’s most valuable technology companies.
In practice, the same 8 people who run the for-profit company also govern the nonprofit that supposedly oversees it.
The result: The Foundation’s “oversight” of OpenAI consists of the company’s leadership overseeing themselves.
Class N Shares: The Governance Mechanism
Section titled “Class N Shares: The Governance Mechanism”What Class N Shares Actually Do
Section titled “What Class N Shares Actually Do”| Asset Type | What It Is | What It Does |
|---|---|---|
| Class N Common Stock | Special governance shares | Exclusive power to appoint/remove all PBC board members |
| 26% Equity Stake | Ordinary shares worth ≈$130B | Financial interest only (dividends, sale proceeds) |
These are legally separate. The Foundation could theoretically:
- Sell its entire 26% equity stake → Still controls the company via Class N
- Give away the equity to charity → Still controls the company via Class N
- Have the equity drop to zero → Still controls the company via Class N
Why This Design?
Section titled “Why This Design?”OpenAI’s lawyers designed this structure to:
| Goal | How Class N Achieves It |
|---|---|
| Prevent hostile takeovers | Can’t buy governance—only Class N holders can appoint board |
| Enable capital raising | Investors get economic upside without governance power |
| Prepare for IPO | Public shareholders buy cash flows, not control |
| Maintain “nonprofit control” narrative | Foundation technically controls company |
The structure is borrowed from dual-class stock designs at Google, Meta, and Snap—but adapted to a nonprofit wrapper.
Who Holds Class N Shares?
Section titled “Who Holds Class N Shares?”Only the OpenAI Foundation (the nonprofit entity) can hold Class N shares. Not Sam Altman personally. Not individual board members. The organization holds them.
Class N Shares ──held by──► OpenAI Foundation (entity) │ governed by │ ▼ Foundation Board (8 people) │ 7 of whom also serve on │ ▼ OpenAI Group PBC BoardCould the board give them away? The Foundation board could theoretically relinquish Class N shares, but:
- They have no incentive to do so
- It would require board approval (same people benefiting from control)
- CA AG oversight creates some friction
The structure ensures the nonprofit cannot be forced to give up control—only the nonprofit can choose to relinquish it.
The Same-Board Problem
Section titled “The Same-Board Problem”Current Board Overlap
Section titled “Current Board Overlap”| Member | Foundation Board | PBC Board | Voting on Both? |
|---|---|---|---|
| Bret Taylor (Chair) | ✓ | ✓ | Yes |
| Sam Altman (CEO) | ✓ | ✓ | Yes |
| Adam D’Angelo | ✓ | ✓ | Yes |
| Sue Desmond-Hellmann | ✓ | ✓ | Yes |
| Paul Nakasone | ✓ | ✓ | Yes |
| Adebayo Ogunlesi | ✓ | ✓ | Yes |
| Nicole Seligman | ✓ | ✓ | Yes |
| Zico Kolter | ✓ | Non-voting observer | No |
7 of 8 Foundation board members have voting power over both the nonprofit mission AND the for-profit company.
What “Oversight” Looks Like
Section titled “What “Oversight” Looks Like”The Foundation’s governance power is real in a legal sense—but exercised by people with strong incentives to prioritize the company’s commercial success.
External vs. Internal Capture
Section titled “External vs. Internal Capture”| Risk Type | Definition | Structure’s Protection | Effectiveness |
|---|---|---|---|
| External capture | Hostile shareholders take control | Class N shares can’t be bought; governance stays with Foundation | Strong |
| Internal capture | Board prioritizes profit over mission | None—same people control both entities | None |
The structure is clever legal engineering against outsiders. It does nothing against insiders.
Why No Charity-Minded Directors?
Section titled “Why No Charity-Minded Directors?”The Selection Problem
Section titled “The Selection Problem”If you put genuinely charity-minded people on the Foundation board—people who take “beneficial AI for humanity” literally—they might:
| Action | Problem for OpenAI |
|---|---|
| Sell the $130B stake and donate it | Loses the asset, creates pressure to actually do philanthropy |
| Block profit-maximizing decisions | ”This deployment timeline is unsafe, we’re using our veto” |
| Demand safety over speed | ”Slow down development until interpretability catches up” |
| Question racing dynamics | ”Is building AGI fastest actually beneficial?” |
| Exercise real oversight | Ask hard questions the PBC board doesn’t want asked |
Current Board Composition
Section titled “Current Board Composition”| Board Member | Background | Charity-Minded? |
|---|---|---|
| Bret Taylor | Salesforce CEO, tech executive | No—business mindset |
| Adam D’Angelo | Quora CEO, Facebook early employee | No—tech founder mindset |
| Sue Desmond-Hellmann | Gates Foundation, Genentech CEO | Maybe—but pharma/tech background |
| Zico Kolter | CMU professor, AI researcher | Maybe—but academic, not activist |
| Paul Nakasone | NSA director, general | No—national security mindset |
| Adebayo Ogunlesi | Private equity, infrastructure investing | No—finance mindset |
| Nicole Seligman | Sony legal, Jenner & Block | No—corporate lawyer |
| Sam Altman | OpenAI CEO | Obviously not |
Zero effective altruists. Zero nonprofit veterans whose primary identity is “I give away money to help humanity.”
The Implicit Selection Criteria
Section titled “The Implicit Selection Criteria”A genuinely charity-minded board might look at the $130B stake and say:
“We should liquidate this over 10 years and fund AI safety research at 100 independent organizations.”
The current board will never do this. The structure isn’t a bug—it’s a feature.
The Signaling Game
Section titled “The Signaling Game”Board Members’ Incentive to Signal Business Focus
Section titled “Board Members’ Incentive to Signal Business Focus”If you’re a board member sitting on both the Foundation and the PBC, your personal interests require signaling:
| Interest | Required Signal |
|---|---|
| Stock appreciation | ”I’m business-focused, growth-oriented” |
| Investor confidence | ”The nonprofit won’t interfere with operations” |
| Employee retention | ”Your equity will be worth something” |
| IPO success | ”This is a real company, not a charity project” |
| Future board seats | ”I’m a serious business person” |
What you do NOT want to signal: “I take the charitable mission so seriously that I might block profitable decisions.”
Observable Signaling Behavior
Section titled “Observable Signaling Behavior”| Signal | What It Tells Investors |
|---|---|
| Board full of CEOs, generals, private equity | ”These are business operators, not philanthropists” |
| Bret Taylor at Davos talking about “AI bubble” risks | ”I think like a finance person, not a missionary” |
| Emphasizing “Public Benefit Corporation” over “nonprofit-controlled" | "We’re a company that does good, not a charity that runs a company” |
| $50M initial donation (0.04% of assets) | “Philanthropy is a footnote, not the focus” |
| RSP loosened before major releases | ”Commercial timelines matter more than safety theater” |
The Quiet Understanding
Section titled “The Quiet Understanding”There’s likely an unspoken agreement among board members:
“We all know the Foundation technically controls the company. We all know we’re supposed to prioritize the mission. But we also all know that if anyone actually does that, they’ll tank the stock, anger investors, and never serve on another board again. So let’s just… not.”
This is reinforced by:
- Social proof: Everyone else on the board is business-minded
- Selection bias: Charity-minded people weren’t invited
- Career incentives: Being “difficult” about mission = reputation damage
- Financial incentives: Everyone benefits from stock appreciation
The Paradox Summarized
Section titled “The Paradox Summarized”The Foundation’s control is supposed to be the point—the thing that makes OpenAI different, the reason the nonprofit structure was preserved.
But everyone involved is incentivized to signal that the control is nominal, won’t be exercised, and shouldn’t worry investors.
The structure exists to say “a nonprofit controls this company.” The people exist to say “but don’t worry, we won’t act like it.”
Post-IPO: Nonprofit-Controlled Public Company
Section titled “Post-IPO: Nonprofit-Controlled Public Company”What Public Shareholders Will Buy
Section titled “What Public Shareholders Will Buy”When OpenAI goes public (projected 2026-2027), the structure will remain:
| Stakeholder | What They Get | What They Don’t Get |
|---|---|---|
| Public shareholders | Economic exposure, dividends, stock appreciation | Board voting rights, governance say |
| Microsoft (27%) | Same as public—money, no votes | Board seats, veto power |
| Foundation (26%) | Money AND 100% governance control | Nothing—they have everything |
Precedents for Dual-Class Structures
Section titled “Precedents for Dual-Class Structures”| Company | Control Mechanism | Public Shareholder Voting |
|---|---|---|
| Meta | Zuckerberg holds 55%+ voting via supervoting shares | Limited |
| Founders hold Class B (10x voting power) | Limited | |
| Snap | Public shares have zero votes | None |
| OpenAI | Nonprofit holds Class N (100% board control) | None |
OpenAI is more extreme—the controlling entity is a nonprofit with a stated mission that could theoretically conflict with shareholder value.
Why Investors Might Accept This
Section titled “Why Investors Might Accept This”- They already did: SoftBank, Microsoft, Thrive invested billions with no governance rights
- Growth story: If you believe OpenAI will 10x, governance doesn’t matter
- Trust in management: Investors bet on Altman, not the structure
- Precedent: Tech dual-class structures are normalized
Why This Is Unprecedented
Section titled “Why This Is Unprecedented”| Normal Dual-Class | OpenAI Structure |
|---|---|
| Founder controls company | Nonprofit controls company |
| Founder’s interest = stock price | Foundation’s interest = “beneficial AI” (undefined) |
| Founder can be sued for fiduciary breach to shareholders | Foundation’s fiduciary duty is to mission, not shareholders |
| Clear profit motive | PBC must balance profit with public benefit |
The pitch to public investors:
“Buy our stock. A nonprofit with a vague mission controls the company. The nonprofit board is mostly the same people running the company. Trust us.”
Potential Post-IPO Complications
Section titled “Potential Post-IPO Complications”| Issue | Risk Level |
|---|---|
| Proxy advisors (ISS, Glass Lewis) recommend against | Medium |
| Institutional investors demand governance reforms | Medium |
| SEC scrutiny of nonprofit control disclosure | Low-Medium |
| Investor lawsuits if Foundation decisions hurt stock | High (post-IPO) |
| Index fund eligibility affected by governance | Low |
The likely outcome: OpenAI IPOs anyway, investors buy anyway because AI hype, and the governance weirdness only matters if something goes wrong.
What Would Real Accountability Look Like?
Section titled “What Would Real Accountability Look Like?”Anthropic’s Alternative: Long-Term Benefit Trust
Section titled “Anthropic’s Alternative: Long-Term Benefit Trust”AnthropicLabAnthropicComprehensive profile of Anthropic, founded in 2021 by seven former OpenAI researchers (Dario and Daniela Amodei, Chris Olah, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish) with early funding...Quality: 51/100’s Long-Term Benefit TrustLong Term Benefit TrustAnthropic's Long-Term Benefit Trust represents an innovative but potentially limited governance mechanism where financially disinterested trustees can appoint board members to balance public benefi...Quality: 70/100 attempts to solve the same problem differently:
| Aspect | OpenAI Foundation | Anthropic LTBT |
|---|---|---|
| Control mechanism | Class N shares (direct) | Shareholder pledge (contractual) |
| Financial stake | $130B equity (26%) | None—pure governance |
| Board overlap | 7 of 8 overlap with company | Designed to be independent |
| Conflict of interest | Foundation benefits from stock appreciation | Trust has no financial stake |
Anthropic’s structure attempts to separate governance from financial interest. OpenAI’s structure combines them in the same people.
What A Better OpenAI Structure Could Look Like
Section titled “What A Better OpenAI Structure Could Look Like”| Element | Current | Better Design |
|---|---|---|
| Class N holders | Foundation board (same as company) | Fully independent trustees |
| Trustee qualifications | Tech/finance executives | Nonprofit leaders, ethicists, safety researchers |
| Financial stake | Trustees benefit from stock appreciation | Trustees have no OpenAI financial interest |
| Term limits | None | Rotating terms, mandatory refreshment |
| Selection | Self-perpetuating (board picks successors) | Independent nominating process |
OpenAI chose not to build this separation. The current structure maximizes control retention while claiming nonprofit oversight.
Paths Forward: How Could This Improve?
Section titled “Paths Forward: How Could This Improve?”The current situation has no clean exit. The same people who benefit from OpenAI’s commercial success control the entity meant to hold them accountable. Any solution needs to break this link.
Realistic Forcing Functions
Section titled “Realistic Forcing Functions”| Path | Probability | Quality of Outcome |
|---|---|---|
| Status quo continues | 40% | Poor—governance theater persists |
| Incremental pressure yields modest reforms | 35% | Mediocre—slightly better but unchanged fundamentally |
| Major forcing event (lawsuit/scandal) | 20% | Could be good or chaotic |
| Voluntary restructuring to real independence | 5% | Best outcome but least likely |
1. Musk Lawsuit Forces Change
Section titled “1. Musk Lawsuit Forces Change”The Musk lawsuitMusk Openai LawsuitElon Musk is suing OpenAI, Sam Altman, Greg Brockman, and Microsoft for $79-134B, alleging they fraudulently induced his $38M donation by promising to maintain nonprofit status, then converted to f...Quality: 70/100 is currently the primary external forcing function:
| Outcome | What Happens |
|---|---|
| Large judgment | Foundation loses assets; precedent deters similar structures |
| Settlement with governance terms | Musk takes cash + Foundation agrees to reforms |
| OpenAI wins | Validates the structure; status quo preserved |
Irony: The best accountability mechanism may be a billionaire suing to recover his donation—and keeping it.
2. California AG Gets Serious
Section titled “2. California AG Gets Serious”The Attorney General could demand:
- Real board independence (separate people)
- Minimum philanthropic deployment rates
- Independent audits of mission alignment
Why it hasn’t happened: Political constraints, limited resources, OpenAI employs thousands in CA, no precedent at this scale.
3. IPO Creates New Pressure Points
Section titled “3. IPO Creates New Pressure Points”Post-IPO, new tools become available:
- Proxy advisory firms recommend against governance structure
- Institutional investors demand reforms for ESG compliance
- Shareholder proposals create annual votes (even if non-binding)
- Media scrutiny increases with public company disclosure
Limitation: Public shareholders have no actual voting power. Pressure is reputational only.
4. Board Composition Changes Over Time
Section titled “4. Board Composition Changes Over Time”Current board is self-perpetuating—they pick successors. But:
- Members age out, resign (Summers already gone due to Epstein scandal)
- Future scandals could force departures
- External pressure could change selection criteria
Best case: Over 10-15 years, gradual shift toward independent directors. Realistic case: Replacements are similar tech/finance people.
5. Catastrophic Incident Forces Reckoning
Section titled “5. Catastrophic Incident Forces Reckoning”If OpenAI causes or enables serious harm:
- Regulatory intervention becomes politically viable
- Board members face personal liability
- Public pressure demands structural reform
The dark path: Real accountability only comes after something goes badly wrong.
What High-Impact Interventions Could Help
Section titled “What High-Impact Interventions Could Help”| Intervention | Cost | Potential Impact |
|---|---|---|
| Support EyesOnOpenAI coalition | $50-100K | Sustained AG pressure |
| Fund investigative journalism | $100-200K | Document governance failures |
| Academic research on governance | $50-100K | Establish this as case study |
| Prepare for IPO pressure coordination | $200-500K | Institutional investor alignment |
The structure was designed to resist change from within. Change will have to come from external pressure, legal action, or crisis.
Key Takeaways
Section titled “Key Takeaways”-
Class N shares are real legal power—the Foundation genuinely controls board appointments
-
Same-board overlap nullifies the oversight—the people being “overseen” are doing the overseeing
-
Board composition ensures no charity-minded disruption—all business/finance/tech executives
-
Signaling incentives push away from mission—board members are rewarded for prioritizing company over Foundation
-
Post-IPO will be unprecedented—public shareholders buying into a nonprofit-controlled company
-
The structure is governance theater—protects against external capture, enables internal capture
-
This was a choice—OpenAI could have created real independence, as Anthropic attempted
See Also
Section titled “See Also”- OpenAI FoundationOpenai FoundationThe OpenAI Foundation holds 26% equity (~\$130B) in OpenAI Group PBC with governance control, but detailed analysis of board member incentives reveals strong bias toward capital preservation over p...Quality: 87/100 — Assets, spending projections, philanthropic activities
- Musk v. OpenAI LawsuitMusk Openai LawsuitElon Musk is suing OpenAI, Sam Altman, Greg Brockman, and Microsoft for $79-134B, alleging they fraudulently induced his $38M donation by promising to maintain nonprofit status, then converted to f...Quality: 70/100 — Legal challenge to the restructuring
- Long-Term Benefit TrustLong Term Benefit TrustAnthropic's Long-Term Benefit Trust represents an innovative but potentially limited governance mechanism where financially disinterested trustees can appoint board members to balance public benefi...Quality: 70/100 — Anthropic’s contrasting governance approach
- OpenAILabOpenAIComprehensive organizational profile of OpenAI documenting evolution from 2015 non-profit to commercial AGI developer, with detailed analysis of governance crisis, safety researcher exodus (75% of ...Quality: 46/100 — The for-profit company
- AnthropicLabAnthropicComprehensive profile of Anthropic, founded in 2021 by seven former OpenAI researchers (Dario and Daniela Amodei, Chris Olah, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish) with early funding...Quality: 51/100 — Competitor with different governance structure