OpenAI Leadership and Governance

The 2015 founding

OpenAI was announced on December 11, 2015 as a nonprofit AI research lab. The cofounders were Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever (recruited from Google Brain), John Schulman, Wojciech Zaremba, Andrej Karpathy, Pamela Vagata, Trevor Blackwell, and Vicki Cheung. The launch announcement listed an initial $1 billion in pledges from Altman, Musk, Reid Hoffman, Peter Thiel, Y Combinator Research, Microsoft, AWS, and Infosys; in practice, most pledges were never fully called.

The stated mission was to ensure that artificial general intelligence “benefits all of humanity,” with research published openly. The legal vehicle was a 501(c)(3) nonprofit. Both of those framings would shift materially over the following decade — the open-research framing receded after GPT-2's staged release, and the nonprofit-only structure was supplemented by the capped-profit OpenAI LP in 2019 and then replaced outright in October 2025 by the OpenAI Foundation / OpenAI Group PBC arrangement that the 2024 – 2025 conversion fight produced.

Of the original cofounder roster, only Altman, Brockman, and Zaremba were still with OpenAI as of mid-2026 — Altman as CEO and Brockman as President of OpenAI Group PBC, Zaremba on the nonprofit side at the OpenAI Foundation. The rest had left at various points; several had founded competing labs (Musk's xAI, Sutskever's Safe Superintelligence, Schulman's later move to Murati's Thinking Machines Lab); two senior early leaders who joined after the 2015 launch — the Amodeis — left in 2020/2021 to start Anthropic and brought several senior researchers with them (covered on the Anthropic leadership page).

The November 2023 board episode (Nov 17 – 21, 2023)

On Friday, November 17, 2023, the OpenAI board — Helen Toner, Tasha McCauley, Adam D'Angelo, and Ilya Sutskever — announced the firing of CEO Sam Altman, citing a loss of confidence in his communications with the board. Greg Brockman was removed as board chair (but not as President) and resigned from the company in protest the same day. Mira Murati, CTO, was named interim CEO. Within twenty-four hours, Murati was reportedly pushing the board to reinstate Altman.

On Sunday, November 19, the board appointed Emmett Shear (former Twitch CEO) as the second interim CEO, displacing Murati. Microsoft publicly offered to hire Altman, Brockman, and any departing OpenAI employees to lead a new AI research division at Microsoft. Roughly seven hundred of OpenAI's seven hundred and seventy employees signed an open letter threatening to follow Altman to Microsoft if the board did not resign and reinstate him. Sutskever signed the employee letter and publicly recanted his earlier vote.

On Tuesday, November 21 — five days after the firing — Altman returned as CEO. The board was reconstituted with Bret Taylor as chair, Larry Summers as a director, and D'Angelo continuing. (Summers later resigned from the board on November 19, 2025, following the release of his Jeffrey Epstein emails.) Toner, McCauley, and Sutskever were off the board (Sutskever stayed at OpenAI as Chief Scientist for several more months before his May 2024 departure). An independent review by WilmerHale was commissioned and concluded in March 2024 that the prior board's decision did not arise from concerns about product safety, security, financial practices, or statements to investors — but stopped short of detailing what it did arise from.

Public framing in the participants' own voices: Helen Toner has described her account in a TED talk and an academic paper. Sam Altman published a return-day post emphasizing the company's continuity. Greg Brockman published a timeline post with internal screenshots. Contemporaneous reporting in NYT, WSJ, Bloomberg, and The Information remains the most thorough public record. The episode is the loudest single test the OpenAI nonprofit-board structure has been put through and is recurring background to the for-profit conversion fight that followed.

The for-profit conversion fight (2024 – 2026)

OpenAI's structure as of early 2024 had three layers: a 501(c)(3) nonprofit (OpenAI, Inc.), a capped-profit subsidiary controlled by the nonprofit (OpenAI LP) that held the operating assets and contracts, and a contractual relationship with Microsoft tied to the LP. Through 2024 and 2025, the company publicly worked toward removing the cap on returns and restructuring the operating entity as a public-benefit corporation. The transaction closed on October 28, 2025 after roughly a year of negotiation with the California and Delaware Attorneys General.

The closed structure runs as follows. The capped-profit subsidiary became OpenAI Group PBC, a Delaware public-benefit corporation. The renamed nonprofit — the OpenAI Foundation — controls the PBC through special voting rights that let the Foundation appoint and replace every director of the PBC's board, and holds approximately 26% of the PBC's equity (worth roughly $130 billion at closing). Microsoft holds approximately 27% (~$135 billion), with the remainder owned by employees and other investors. The Foundation also holds a warrant that grants it additional equity if the PBC's share price rises more than tenfold over fifteen years, and has committed an initial $25 billion to AI-resilience and healthcare-focused programs.

Multiple parties opposed the conversion in the run-up to closing. Musk v. Altman in the Northern District of California argues the conversion betrays the original founding agreement (covered legally on the lawsuits page). Musk's preliminary-injunction motion asking the court to block the conversion was denied in February 2025; the surviving breach claims went to trial in April 2026 (see the trial row above). The SEIU-affiliated “Coalition for AI Nonprofit Integrity” filed comments with the California Attorney General; a coalition of former OpenAI employees and outside experts filed parallel concerns. State Attorneys General negotiated terms with OpenAI directly through 2024 and 2025; the regulatory clearance preceded the closing.

A further ownership question opened in June 2026. President Trump confirmed on June 5, 2026 that the administration and OpenAI were in talks over a possible US government equity stake — reportedly structured as an OpenAI equity donation (reported at 1–5%) to seed a “Public Wealth Fund,” the mechanism OpenAI itself proposed in its April 2026 policy paper. By early July 2026 the ask had sharpened into a formal proposal to grant the government a 5% stake — roughly $42.6 billion at OpenAI’s ~$852 billion valuation, modeled on the Alaska Permanent Fund and pitched with parallel asks to Anthropic, Google, and Meta — but the proposal remained preliminary and unsigned. A government stake would layer a new actor onto the PBC-plus-Foundation structure if it closes.

The structural questions are now in two places. The legal question of whether the original founding agreement constrained OpenAI's ability to convert has not been answered on the merits: the May 2026 defense verdict held that Musk brought those claims too late, and the appeal to the Ninth Circuit is the only route left to a ruling that could order any part of the conversion unwound — the counts still live in the district court are antitrust-shaped and carry no such remedy. The structural question of whether the post-conversion shape — PBC controlled by an equity-and-warrant-holding nonprofit Foundation — behaves as the original mission required is the open question for the next several years; on paper the Foundation retains director-appointment authority over every Group director and a Safety and Security Committee with scope over all of OpenAI, and it has been adding directors to both boards as a listing approaches (see the July 21, 2026 row).

The 2024 superalignment exodus

The year after the November 2023 board episode produced the most concentrated departure of senior leadership in OpenAI's history. The clustering was pronounced enough that “the exodus” became a single phrase in coverage of the period.

February 2024: Andrej Karpathy left for the second time and later founded Eureka Labs; he joined Anthropic's pretraining research team in May 2026. May 2024: Ilya Sutskever and Jan Leike resigned within days of each other; Leike's public thread stated that “safety culture and processes have taken a backseat to shiny products.” The superalignment team — which had been formed in 2023 with a public commitment of twenty percent of OpenAI's compute budget — was dissolved and its members reassigned. Sutskever later founded Safe Superintelligence Inc.; Leike went to Anthropic.

May–August 2024: Greg Brockman took an extended sabbatical, returning in late August. August 2024: John Schulman, an OpenAI cofounder and the original lead on RLHF and on ChatGPT itself, left for Anthropic and later moved to Mira Murati's startup. September 2024: Mira Murati left and founded Thinking Machines Lab; Bob McGrew (CRO) and Barret Zoph (VP Research) left the same week.

After the recapitalization closed in October 2025, the senior team in the operating PBC consists of Sam Altman (CEO), Greg Brockman (President), Sarah Friar (CFO), and Mark Chen (SVP Research / Chief Research Officer). Denise Dresser, the former Slack CEO, joined as Chief Revenue Officer in the spring-2026 reshuffle. Brad Lightcap — long-time COO — transitioned out of the COO role in spring 2026 to lead special projects reporting directly to Altman. After Fidji Simo stepped down from the CEO-of-Applications role in July 2026, Brockman absorbed her ChatGPT-product, enterprise, and go-to-market portfolio, concentrating the operating business under the President ahead of a possible public listing. Bret Taylor chairs both boards — the OpenAI Foundation, which through special voting rights appoints every director of OpenAI Group PBC, and the Group board itself; the other Foundation directors named in OpenAI's structure page are Adam D'Angelo, Sue Desmond-Hellmann, Zico Kolter, Paul Nakasone, Adebayo Ogunlesi, Nicole Seligman, and Altman, joined in July 2026 by Nubank's David Vélez and BNY's Robin Vince. The current roster should be re-verified at every refresh given the volatility of the prior period.

Spring 2026 produced a second cluster of departures. Kevin Weil, the former CPO who had moved to lead OpenAI for Science, left in April 2026; Bill Peebles (head of the now-shuttered Sora team) and Srinivas Narayanan (B2B Applications) left the same week. Fidji Simo, the CEO of Applications hired from Instacart, took medical leave in April for a neuroimmune condition and stepped down from the full-time role on July 9, 2026 to a part-time advisory position; Kate Rouch, the marketing chief, stepped down for cancer recovery. Barret Zoph — the former VP Research who had left in the September 2024 cluster for Thinking Machines Lab and then rejoined OpenAI in January 2026 to lead enterprise AI sales — left again on June 19, 2026 after five months, with no public reason given. Coverage at the time framed the departures as part of an internal pullback from “side quests” (Sora, OpenAI for Science) ahead of a possible IPO.

The IPO ramp-up also brought senior hires. In June 2026, OpenAI recruited Noam Shazeer — the Transformer co-inventor and Google DeepMind’s Gemini co-lead — as Lead for Architecture Research, and Dean Ball, a former White House OSTP AI-policy adviser, to lead a new Strategic Futures team (frontier-AI policy and internal governance, reporting to Chief Strategy Officer Jason Kwon), starting July 6 (see the June 18, 2026 timeline row). Later in June, OpenAI also poached Paul Meade, Apple’s VP of hardware engineering and the head of its Vision Pro and smart-glasses hardware, to work on OpenAI’s consumer-hardware devices unit. The Shazeer move was the highest-profile inbound transfer from a rival frontier lab since the 2024 exodus ran the other direction.

The safety-leadership churn continued into July 2026. Joshua Achiam — a near-nine-year veteran who had led the Mission Alignment team (the superalignment team's successor, disbanded in February 2026) and then held the “chief futurist” title — told colleagues on July 1 that he would leave the company on July 24. Days later, on July 11, Johannes Heidecke, head of safety systems since 2024, announced his own departure as OpenAI merged its safety and research organizations: safety teams now report to Mia Glaese, elevated to a newly created VP of Research and Safety role, with Saachi Jain as interim head of safety systems. It was the second time in under two years that OpenAI folded its safety organization into a research-reporting structure; critics argued that safety embedded inside research has less independence to delay or block a launch, while CRO Mark Chen framed the change as giving safety “an earlier and more direct role” in model and launch decisions.

The pattern of safety-focused researchers leaving for Anthropic over stated process disagreements echoes the original 2021 founding of Anthropic itself; together they form a multi-year reshuffle covered on the Anthropic leadership page.

The Microsoft entanglement

The Microsoft relationship is the financial structure inside which essentially every other OpenAI governance question has played out. The 2019 commitment ($1 billion) made Azure the exclusive cloud provider. The 2023 expansion (reportedly $10 billion) deepened the financial link and added Bing, Microsoft 365, and the Azure OpenAI Service as strategic-distribution surfaces. The relationship was rewritten twice in the most recent cycle: a definitive agreement at the October 28, 2025 recapitalization and an amendment on April 27, 2026.

Where the structure now stands. Microsoft holds an investment in OpenAI Group PBC valued at approximately $135 billion at closing — roughly 27% on an as-converted diluted basis — making it the largest non-Foundation shareholder. The capped-profit cap is gone, replaced by ordinary equity. Microsoft's IP rights to OpenAI models and products extend through 2032 and now cover post-AGI models with safety guardrails; the IP license was made non-exclusive in the April 2026 amendment, and OpenAI can now serve products on any cloud provider while Azure remains the primary first-launch surface for new product capabilities. OpenAI committed to purchase an additional $250 billion of Azure services as part of the October 2025 deal.

The AGI clause — long the most-watched provision of the contract — survives the restructuring but in modified form. Under the October 2025 agreement, OpenAI's board still declares AGI, but that declaration is now verified by an independent expert panel before the contractual consequences attach. Microsoft's IP rights to OpenAI's research methods (the confidential techniques used to build models and systems) end at the earlier of the expert panel's verification or 2030; Microsoft's IP rights to model weights, architecture, inference code, fine-tuning code, and data-center hardware/software are excluded from research-IP and continue independently. Microsoft can independently pursue AGI alone or with third parties, including using OpenAI IP up to defined compute thresholds set well above today's frontier-model scale.

The April 2026 amendment further loosened the commercial exclusivity. Microsoft no longer pays a revenue share to OpenAI; the OpenAI-to-Microsoft revenue share continues through 2030 at the same percentage but is now subject to a total cap. OpenAI can release open-weight models that meet capability criteria and can provide API access to U.S. government national-security customers regardless of the cloud provider. Together, the two 2025 – 2026 rewrites convert the relationship from an exclusive bilateral structure into something closer to a major shareholder plus preferred cloud partner.

The November 2023 board episode briefly tested the older structure — Microsoft offered to hire Altman, Brockman, and any departing OpenAI employees, demonstrating that the operating-talent risk was concentrated in a small group whose loyalty was as much to Altman as to the corporate vehicle. Whether the new PBC-plus-equity-stake structure substantively addresses that entanglement or only changes its legal form is a question the Musk v. Altman trial left open — the May 2026 verdict turned on the statute of limitations rather than the merits (see the trial row above), and the Ninth Circuit appeal is where it goes next.

Governance comparison — OpenAI vs. Anthropic vs. xAI vs. DeepMind

The four major frontier-model labs sit on four meaningfully different governance shapes. The differences are load-bearing for how each company can be expected to behave under stress.

OpenAI spent its first decade as a nonprofit (OpenAI, Inc.) controlling a capped-profit subsidiary (OpenAI LP). That arrangement ended on October 28, 2025. Per OpenAI's own structure page, the nonprofit is now the OpenAI Foundation and the operating company is OpenAI Group PBC, a Delaware public-benefit corporation; the cap on investor returns is gone and every holder now owns ordinary stock. The Foundation controls the PBC through special voting rights that let it appoint and replace every Group director, holds roughly 26% of the equity plus a tenfold-share-price warrant, and keeps the Safety and Security Committee at the Foundation level with scope over all of OpenAI. Mission authority therefore sits with a controlling nonprofit board rather than with an independent, financially-disinterested trust — and the November 2023 board episode is the one real-world test of what that authority is worth under pressure: the firing was effective, and it did not survive the employee response and the Microsoft offer.

Anthropic, PBC is a Delaware public benefit corporation with the Long-Term Benefit Trust (LTBT) above the board — an independent body of financially-disinterested trustees whose right to elect directors ramps in tranches tied to cumulative equity raised. That ramp is no longer prospective: with the April 2026 appointment of Vas Narasimhan, Anthropic said trust-appointed directors reached a majority of the board for the first time. The structure is designed to make the company harder to deflect from its stated mission as the financial stakes grow. (See the LTBT explainer on the Anthropic leadership page.)

xAI is the outlier in the other direction: no PBC form, no independent body equivalent to the LTBT, and no longer a standalone company. Per Nevada Secretary of State filings, xAI was briefly amended in April 2023 to Nevada Benefit Corporation form (the state's PBC analog), then terminated that benefit-corporation status in May 2024 and returned to standard for-profit corporation form. The merger chain since then — the March 2025 X merger into xAI Holdings, then the February 2026 SpaceX merger that put SpaceX on top — concentrated voting control through Elon Musk's personal cap-table position rather than dispersing it, and the AI lab now operates as a division inside SpaceX, which listed on Nasdaq (SPCX) in June 2026. Control of the lab consequently runs up through a public company's cap table to Musk's controlling-shareholder position in it. (See the Musk-personal-control governance section on the xAI leadership page for the full detail.)

Google DeepMind is a wholly-owned subsidiary of Alphabet, governed by ordinary corporate-board mechanics. Mission framing and ethics review run through internal Alphabet processes rather than an independent external structure. The lab's incentives ultimately answer to Alphabet's public-company shareholders. All four shapes are now settled rather than in transition — the open question has shifted from what structure to what public markets do to it: OpenAI and Anthropic each filed a confidential draft S-1 in mid-2026, DeepMind has answered to public shareholders all along, and xAI already sits inside a listed parent.

The public voice

OpenAI has historically had multiple high-visibility executive voices in parallel — Sam Altman, Greg Brockman, Mira Murati, Ilya Sutskever, John Schulman, and others have all had distinct public profiles at different points. The pattern is more diffuse than Anthropic's deliberately-narrow concentration around Dario Amodei and Jack Clark (covered on the Anthropic leadership page).

The diffusion has costs and benefits. The November 2023 episode and the 2024 exodus both produced multiple competing public framings of the same events; coverage was unusually rich because of it. At the same time, the multiple-voices pattern is part of why every governance episode at OpenAI plays out in public for weeks — departed executives publish their own framing, current executives respond, contemporaneous reporting captures the back-and-forth.