The clause.
For six years, the most consequential sentence in artificial intelligence was not a line of model output. It was a contract provision: the clause in the 2019 Microsoft–OpenAI agreement that said once OpenAI achieved AGI, Microsoft’s access to the technology would end.
The clause was written to protect a mission. OpenAI’s founding promise was that artificial general intelligence should benefit humanity, not be captured by a single corporation — so the contract walled AGI off from the commercial partner. The provision restricted Microsoft’s access to OpenAI’s technology once AGI was achieved, and it gave OpenAI’s board the unilateral authority to declare that AGI had arrived.
There was a problem buried in it. The contract contained no agreed definition of AGI. The working language — systems that “surpass humans in most economically valuable work,” paired in some accounts with a threshold of $100 billion in potential profits — was a description, not a test. There was no regulatory body to certify AGI, no objective milestone, no timer. The clause was, in one widely-quoted framing, a time bomb without a timer: detonation tied not to a measurable event but to OpenAI’s own interpretation of when the moment had come.
Then the clause became the obstacle to everything OpenAI wanted to do next. To restructure into a public benefit corporation, raise the capital its compute bills required, and clear a path toward an eventual public offering, OpenAI needed Microsoft’s sign-off — and Microsoft’s single largest point of leverage was the very clause that threatened to strand its investment. The mission-protection provision had become the price of the mission’s commercialization.
The structural argument I want to make: the AGI clause is the cleanest case study available of what happens when a contractual definition of AGI collides with the capital structure built on top of it — and the capital structure wins. The clause did not survive contact with a $500 billion recapitalization. It was not triggered, litigated, or vindicated. It was renegotiated into a verification step that no longer ends the relationship — and the thing it was written to govern, the arrival of AGI, was quietly converted from a cliff into a procedure.
The headline integrative finding: Across two amendments — October 28, 2025 and April 27, 2026 — the clause was systematically defused. Unilateral declaration became panel verification. A trigger that ended Microsoft’s access became a milestone that ends only some of it, and later. A provision that could have escalated payments was decoupled from payments entirely. By the end, “AGI” in the contract had been demoted from the event that severs the partnership to an administrative checkpoint that the partnership is structured to survive. The mission language remains in the documents. Its teeth do not.
This essay walks the clause as originally written, why it became the binding constraint on OpenAI’s restructuring, the October 2025 recapitalization that began defusing it, the April 2026 amendment that finished the job, what “AGI verification” actually means now, who won and who conceded, and the structural reading of what it means that the most important definition in AI turned out to be a negotiable contract term.
The clause.
How a contractual
definition of AGI met
the capital built
on top of it.
clause stood in the way of
post-AGI models · the clause reversed
payments decoupled from AGI
OpenAI models live on AWS Bedrock
fireable without
catastrophic cost
to the firer
A provision written to wall AGI off from a single corporation became the price of that corporation’s continued partnership — renegotiated from a unilateral, deal-ending trigger into a jointly-verified, consequence-free checkpoint. The form of the mission survived; its force was traded for the capital the restructuring required.Thorsten Meyer · The Clause · AI Governance 03
By Thorsten Meyer — May 2026
This is a dispatch in the AI Governance & Corporate Structure track. Earlier pieces walked the Musk v. Altman verdict and the structural mirror of Anthropic’s cap table. This one is about a single clause — the AGI provision in the Microsoft–OpenAI contract — because the way it was written, fought over, and ultimately dissolved is the most legible example we have of how the governance ideals embedded in AI’s founding documents fare when they meet the capital required to build the technology.
The structural argument I want to make: a definition of AGI written into a contract is only as durable as the commercial relationship it constrains. The clause was real, binding, and genuinely consequential — Microsoft’s internal urgency to resolve it was reportedly extreme — and that is precisely why it could not survive. A provision that powerful, standing between OpenAI and the capital it needed, was not going to be left to the discretion of a board declaration and an undefined term. It was going to be negotiated into something both parties could live with, and “something both parties can live with” is, by construction, not a provision that ends the relationship.
The headline integrative finding: The clause’s journey from “doomsday provision” to “verification step” tracks the broader trajectory of AI’s founding governance ideals under capital pressure. The mission-protective intent survives in form — there is still a panel, still a verification, still language about AGI benefiting humanity. The mission-protective force is gone — verification no longer ends the partnership, no longer triggers payments, and no longer rests with OpenAI alone. This is not a scandal; both sides got something they needed. It is, instead, a clean demonstration that when a governance mechanism and a capital structure are in tension, the governance mechanism is what bends.
This essay walks the clause as written (Section I), why it bound OpenAI’s restructuring (Section II), the October 2025 recapitalization (Section III), the April 2026 amendment (Section IV), what AGI verification means now (Section V), who won and who conceded (Section VI), and the structural reading of a negotiable definition (Section VII).
LEAN PROGRAMMING FOR FORMAL SOFTWARE VERIFICATION: Mathematical proof systems and logical frameworks for verified computation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
I · The clause as written · a definition with no definition
The provision crystallization. To understand the renegotiation, you have to understand exactly how much the original clause tried to do, and how little it actually specified.
What the clause said
The trigger: once OpenAI achieved AGI, the special terms of the Microsoft relationship would change — Microsoft’s access to OpenAI’s most advanced technology would be restricted, the IP license would not extend to post-AGI systems, and the commercial arrangement built around pre-AGI models would not carry forward.
The declaration power: the authority to declare that AGI had been reached sat with OpenAI’s board. Not with a regulator, not with a joint body, not with an objective test. OpenAI, unilaterally, could decide that the triggering event had occurred.
The working definition: AGI described as systems that surpass humans in most economically valuable work, paired in some accounts with a profit threshold — potential profits exceeding $100 billion — as a rough marker of the scale at which the question became live.
The hole in the middle
No agreed test: the contract did not define, operationally, what would count as AGI. Surpassing humans at “most economically valuable work” is a sentence, not a measurement. There is no benchmark, no certifying authority, no threshold that both parties had agreed in advance would settle the question.
The asymmetry this created: because the trigger was undefined and the declaration was unilateral, the clause concentrated enormous discretionary power in OpenAI and enormous exposure in Microsoft. Microsoft had built Azure’s AI offering, Copilot across Windows and Office and GitHub, and a substantial part of its forward strategy on access to OpenAI’s models — and that access could, in principle, be curtailed by a board vote on a term nobody had defined.
Why it was written this way
The mission logic: in 2019, the clause made sense as mission protection. If OpenAI genuinely believed AGI could arrive and could be dangerous if captured, then walling it off from the commercial partner — and keeping the declaration in mission-aligned hands — was a coherent way to encode “this technology is different and should not simply become Microsoft’s product.”
The latent contradiction: but the same provision that protected the mission also made OpenAI’s commercial relationship fundamentally unstable, because the partner’s access rested on an undefined term controlled by the other side. A clause coherent as mission protection was incoherent as the foundation for the largest commercial partnership in technology — and that contradiction is what eventually had to be resolved.
The clause observation
The AGI clause tried to encode a governance ideal — AGI should not be captured by one company — into an enforceable contract term, and it did so with a trigger nobody had defined and a declaration power held by one side. That structure was powerful precisely because it was undefined and unilateral. And it was unsustainable for exactly the same reason: a partnership cannot be built on a foundation either party can redefine and detonate at will. The clause was a governance ideal that doubled as a commercial time bomb.
Principles of Agentic AI Governance: A Playbook for Managing AI Risk, Fairness, and Compliance (Agentic Governance and Architecture)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
II · Why it bound the restructuring · the clause as the gate
The leverage crystallization. The clause sat dormant as an abstract risk for years. What made it the central drama of 2025 was OpenAI’s need to restructure — because the restructuring ran directly through Microsoft’s consent, and the clause was Microsoft’s leverage.
What OpenAI needed
The PBC conversion: OpenAI needed to convert its for-profit arm into a public benefit corporation controlled by the nonprofit foundation — the structure required to raise capital at the scale its compute commitments demanded and to keep a credible path toward an eventual public offering.
The capital: OpenAI’s inference and training costs require capital on a scale that the original nonprofit-with-a-capped-profit-subsidiary structure could not efficiently raise. The recapitalization was not optional; it was the financial precondition for continuing to operate at frontier scale.
Microsoft’s consent: the restructuring required Microsoft’s agreement, because Microsoft’s existing rights — to IP, to revenue share, to exclusivity — were embedded in the very agreement being rewritten. Microsoft could not be routed around. It had to be negotiated with.
Why the clause was the gate
Microsoft’s exposure ran through the clause: the AGI provision was the single term that most threatened Microsoft’s investment, because it could strand everything Microsoft had built. Resolving the restructuring meant resolving the clause — they were the same negotiation.
The mutual hostage situation: OpenAI held the declaration power (it could, in theory, declare “sufficient AGI” and limit Microsoft’s access). Microsoft held consent power (it could decline to approve the restructuring OpenAI needed). Each side had a weapon pointed at the other, and neither weapon could be fired without catastrophic cost to the firer. OpenAI declaring AGI to escape Microsoft would invite regulatory scrutiny and blow up its most important commercial relationship; Microsoft blocking the restructuring would damage the company whose technology underpins its AI strategy.
The boiling point
The 2025 tension: negotiations reportedly reached a boiling point over the summer of 2025, with OpenAI said to have considered raising antitrust concerns about Microsoft’s conduct, and Microsoft’s internal urgency to resolve the clause reportedly assessed as extremely high. The relationship had moved from symbiosis toward open rivalry while remaining mutually indispensable.
The diversification signals: in the same window, Microsoft integrated Anthropic’s Claude models into Office 365 Copilot, and OpenAI signed a massive Oracle cloud commitment and prepared a path to additional cloud providers. Both sides were visibly reducing dependence on the other — which is exactly the posture that makes a negotiated resolution possible, because each side’s best alternative to agreement was improving.
The gate observation
The clause became the binding constraint on OpenAI’s restructuring because the restructuring required Microsoft’s consent, and the clause was the term on which Microsoft’s consent most depended. The mission-protective provision had become the lever Microsoft could pull to extract concessions. The mutual-hostage structure guaranteed that the clause would be renegotiated rather than triggered — because triggering it, in either direction, was ruinous, while renegotiating it let both sides convert their weapons into terms. The only question was what the clause would become.
MASTERING WEB3 WITH AI AGENTS: BUILDING AUTONOMOUS BLOCKCHAIN APPLICATIONS WITH SMART CONTRACTS, AI AUTOMATION, DEFI PROTOCOLS, AND DECENTRALIZED INTELLIGENCE
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
III · The October 2025 recapitalization · the clause begins to dissolve
The first-amendment crystallization. On October 28, 2025, the companies announced the restructured agreement that completed OpenAI’s conversion to a public benefit corporation — and began converting the AGI clause from a trigger into a procedure.
The corporate restructure
The new entity: OpenAI’s for-profit arm became OpenAI Group PBC, a public benefit corporation. The nonprofit OpenAI Foundation retained control, holding roughly a 26% stake valued around $130 billion, with current and former employees and investors owning the remainder. The Group was valued at roughly $500 billion in the round.
Microsoft’s position: Microsoft’s stake was set at approximately 27% on an as-converted diluted basis, valued around $135 billion — down from a 32.5% stake on the prior basis, but a substantial return on its roughly $13.8 billion invested. In exchange, OpenAI contracted to purchase an incremental $250 billion in Azure services.
What happened to the clause
Verification replaces unilateral declaration: the single most important change. The power to declare AGI no longer sat with OpenAI’s board alone. Any AGI declaration would now be verified by an independent panel of experts, jointly established by both companies. The unilateral trigger — the feature that made the clause a time bomb — was removed. No single entity could now define when, or whether, AGI had been achieved.
IP rights extended through and past AGI: Microsoft’s IP rights for models and products were extended through 2032 — and, critically, now included models developed after an AGI declaration, under safety guardrails. The clause’s original core effect — that AGI ends Microsoft’s access — was substantially reversed: Microsoft’s access now explicitly survives AGI, through 2032. Its rights to OpenAI’s confidential research methods were set to last until AGI verification or 2030, whichever came first.
Exclusivity loosened: Microsoft gave up its right of first refusal to be OpenAI’s compute provider. OpenAI gained the freedom to jointly develop some products with third parties; co-developed API products remained Azure-exclusive, while non-API products could run on any cloud.
What survived
The revenue share, until verified AGI: the revenue-sharing arrangement remained in place until the expert panel verifies AGI, though payments were to be made over a longer period. The mission language and the panel survived. The cliff did not.
The recapitalization observation
The October 2025 amendment did the heavy structural work: it converted OpenAI into a PBC, set Microsoft’s stake and Azure commitment, and — decisively — replaced unilateral AGI declaration with independent-panel verification while extending Microsoft’s access through and beyond AGI to 2032. In a single step, the clause went from “AGI ends Microsoft’s access, on OpenAI’s say-so” to “AGI is verified by a joint panel, and Microsoft’s access continues regardless through 2032.” The trigger had become a checkpoint. The cliff had become a procedure. What remained was to sever the last link between that procedure and the money.
Bryzova Smart Ring | No Subscription, Heart Rate & Blood Oxygen Monitoring, Sleep & Women’s Health Tracking, AI Health Advisor, IP68 Waterproof Fitness Ring for Android & iOS (Silver, 8#)
All-Day Wellness Tracking: This ring automatically monitors your heart rate and skin temperature every 30 minutes, along with…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
IV · The April 2026 amendment · the clause finishes dying
The second-amendment crystallization. On April 27, 2026, the companies amended the partnership again — and the speed of what followed told the real story.
The amendment
Exclusivity ends: the April amendment ended Microsoft’s status as the sole authorized third-party cloud provider for OpenAI’s frontier models. The next day — not the next week, the next day — OpenAI’s models appeared on AWS Bedrock. The immediacy demonstrated that the infrastructure to multi-source had been ready and waiting; only the contract term was holding it back.
The Azure first-shot clause: Azure retained a softened priority — OpenAI products would still launch first on Azure “unless Microsoft cannot and chooses not to support the necessary capabilities” — a first-shot right rather than an exclusivity.
What happened to the clause’s last teeth
Payments decoupled from AGI: under the renegotiated terms, the revenue-share obligation was capped and decoupled from AGI milestones entirely. The original structure, under which AGI-related provisions could have escalated or altered payments, was dismantled. AGI verification still ends Microsoft’s research-IP rights — but it no longer triggers any revenue-share escalation, and no longer determines whether Microsoft’s license continues. OpenAI was reported to save on the order of $97 billion in payments to Microsoft through 2030 under the renegotiated terms — the financial measure of how much the old clause-laden structure had been worth, and how much dismantling it returned to OpenAI.
Verification persists, but inert: the April amendment did not remove the independent-panel verification mechanism introduced in October. It removed the verification’s consequences. Verifying AGI no longer governs whether the license continues. The panel remains; what it decides no longer detonates anything.
The arc completed
From the original clause to here: the 2019 clause said AGI (declared unilaterally by OpenAI, undefined) ends Microsoft’s access and unwinds the deal. The April 2026 reality says AGI (verified by a joint panel, still undefined operationally) ends only Microsoft’s research-IP rights, changes no payments, and leaves the broad license intact through 2032. Every load-bearing element of the original clause — unilateral declaration, access termination, payment consequences — had been removed across two amendments. What remained was a verification step attached to a narrow consequence.
The amendment observation
The April 2026 amendment finished what October began: it severed AGI verification from money and from the license’s continuation, leaving a mechanism that verifies a milestone without that milestone changing much of anything. The next-day AWS launch proved the exclusivity had been the only real lock, and the $97 billion in projected savings priced the dismantling. The clause that was once a doomsday provision had become an administrative footnote — present in the documents, absent in effect.
V · What “AGI verification” means now · a procedure without a cliff
The mechanism crystallization. It is worth being precise about what survives, because the survival is real and the temptation is to say the clause simply vanished. It did not vanish. It was hollowed.
What the panel is
The structure: an independent panel of experts, jointly established by Microsoft and OpenAI, with the authority to verify any AGI declaration OpenAI makes. It is designed to operate with autonomy and transparency, so that neither party can unilaterally define the moment.
What it fixed: the panel solves the original clause’s worst feature — unilateral declaration of an undefined term. By moving the call to a joint, independent body, it removes the time-bomb dynamic. Neither side can now weaponize the AGI declaration against the other. As a governance mechanism for resolving a genuinely hard definitional question, the panel is a real improvement over a board vote.
What the panel no longer does
It does not end the partnership: verification no longer terminates Microsoft’s broad access. Microsoft’s IP rights for models and products run through 2032, including post-AGI systems under safety guardrails.
It does not move money: verification is decoupled from the revenue share. Confirming AGI does not escalate, trigger, or alter payments.
It governs one narrow thing: verified AGI ends Microsoft’s rights to OpenAI’s confidential research methods (or those rights end in 2030 regardless, whichever comes first). That is the residual consequence — meaningful, but a fraction of the original clause’s reach.
The definitional irony
AGI is still undefined: notably, none of the amendments resolved the underlying problem — what AGI actually is. The operational definition remains as absent as it was in 2019. The parties did not agree on what AGI means. They agreed that whatever it means, its arrival will be verified by a panel and will no longer blow up the deal. They solved the contractual problem (who decides, and what happens) without solving the conceptual one (what is the thing). That is a perfectly rational thing for two companies to do — and a telling one, because it means the most important definition in AI was rendered commercially irrelevant before it was ever pinned down.
The mechanism observation
What survives is a verification procedure that fixed the clause’s worst governance flaw — unilateral declaration — while being stripped of the consequences that made the clause matter. The panel is a genuine improvement as a way to resolve a hard question; it is also, now, a mechanism whose answer changes very little. AGI in this contract has become a thing that will be carefully verified and then will not detonate — which is either mature governance or defanged governance, depending on whether you think the clause’s teeth were a bug or the point.
VI · Who won and who conceded · reading the ledger
The settlement crystallization. It is tempting to read the dissolution of the clause as Microsoft winning — it kept access AGI was supposed to end. The ledger is more balanced than that, and the balance is the point.
What Microsoft got
Durable access: IP rights through 2032, explicitly including post-AGI models — the thing the original clause most threatened. The single largest risk to Microsoft’s AI strategy was removed.
A clean stake: roughly 27% of OpenAI Group PBC, valued around $135 billion, a large return on roughly $13.8 billion invested.
Independent AGI development: Microsoft gained the freedom to pursue AGI on its own or with others — though with compute limits and continued revenue-sharing applying if it develops AGI using OpenAI’s technology before the panel verifies AGI.
What Microsoft conceded
Exclusivity: it gave up right of first refusal on compute and its status as sole frontier-model cloud provider. OpenAI is now multi-cloud, with AWS live.
The revenue cap and timing: payments were capped and decoupled from AGI, and OpenAI’s projected ~$97 billion in savings through 2030 is, in part, Microsoft’s foregone upside.
What OpenAI got
The restructuring it needed: PBC conversion, nonprofit control retained, a path to the capital and eventual public offering its scale requires.
Freedom from the cliff and the cap: the AGI clause no longer threatens to unwind the deal, payments are capped and decoupled, and OpenAI can develop with third parties and across clouds. The ~$97 billion in savings is the financial prize.
What OpenAI conceded
Mission-protective force: the clause that walled AGI off from the commercial partner is gone in substance. Microsoft’s access now survives AGI. The nonprofit retains control of the entity, but the specific contractual mechanism that said “AGI is different and ends the commercial relationship” was traded away for the capital and consent the restructuring required.
The unilateral lever: OpenAI gave up the power to declare AGI on its own terms — a power it could never have used without catastrophe, but a lever nonetheless.
The ledger observation
Both sides won something they needed and conceded something they valued, which is what a settlement between mutually indispensable parties looks like. Microsoft secured durable access and gave up exclusivity and upside; OpenAI secured its restructuring and gave up the mission-protective teeth of the clause. The clean reading is not “Microsoft won” but “the commercial relationship won” — both companies optimized for continuing to do business together, and the casualty was the provision that contemplated not doing business together once AGI arrived. The mission ideal was the thing on the table that neither party, in the end, was willing to let block the deal.
VII · The structural reading · a definition that turned out to be negotiable
The synthesis crystallization. The AGI clause is usually discussed as Microsoft–OpenAI corporate drama. Its real significance is as evidence — the clearest available — of how AI’s founding governance ideals behave when they meet the capital structures built on top of them.
Observation 1 · The clause was governance encoded as contract, and contract is negotiable
The empirical signal: the clause was a genuine attempt to encode “AGI should not be captured by one company” into an enforceable term. It was real and binding. And across two amendments it was negotiated into inertness.
The structural reading: a governance ideal encoded as a contract term inherits the negotiability of a contract term. When the ideal and the capital structure came into tension — when the clause stood between OpenAI and a $500 billion recapitalization — the ideal was what bent, because contracts are precisely the instruments parties rewrite when continuing the relationship is worth more than honoring the original term. The clause was not betrayed or breached. It was renegotiated, which is what contracts are for. That is the lesson: encoding a mission constraint as a contract term makes it binding and negotiable in the same stroke.
Observation 2 · The mutual-hostage structure guaranteed renegotiation, not enforcement
The empirical signal: each side held a weapon (declaration; consent) that was ruinous to fire. Neither fired. Both converted their weapons into terms.
The structural reading: a clause that can only be enforced at catastrophic cost is a clause that will be renegotiated, not enforced. The AGI provision was never going to be triggered, because triggering it — in either direction — destroyed value for both parties. Its function was never to be exercised; it was to be a bargaining position. Mission-protective provisions that depend on a nuclear option tend to be defused at the negotiating table, because the nuclear option’s unusability is exactly what makes it tradeable.
Observation 3 · The hard question was made irrelevant rather than answered
The empirical signal: none of the amendments defined AGI. They built a panel to verify an undefined term and then removed the consequences of verification.
The structural reading: the parties solved the contractual problem without solving the conceptual one — and rendered the conceptual one commercially moot in the process. “What is AGI” remains unanswered; “what happens when someone says we have it” now answers: a panel checks, and not much follows. The most consequential definitional question in the field was routed around rather than resolved, because resolving it was unnecessary once its commercial stakes were removed. The definition did not get clearer. It got less important.
Observation 4 · The form of the mission survives; the force does not
The empirical signal: there is still a nonprofit in control, still a panel, still language about AGI and humanity. There is no longer a clause that ends the partnership when AGI arrives.
The structural reading: the mission’s institutional form was preserved while its specific enforcement mechanism was dismantled — a pattern visible across AI governance, where the structures that signal mission (foundations, charters, boards, panels) persist while the particular levers that would constrain commercial behavior get traded away under capital pressure. Whether that is maturation or capture depends on whether you believe the surviving form still binds. The clause’s history suggests caution: the form survived precisely because it had been emptied of the force that would have made it costly.
What this is not
It is not a claim that anyone acted in bad faith. Both parties negotiated rationally; the panel is a real governance improvement; the settlement was balanced. The point is structural, not moral.
It is not a claim that the mission is dead. The nonprofit retains control; the PBC form imposes some duties; the panel exists. The claim is narrower: the specific clause that encoded “AGI ends the commercial relationship” was dissolved.
It is not unique to OpenAI. The pattern — mission ideals encoded as governance mechanisms, then renegotiated under capital pressure while the form persists — is the central tension of the entire AI-governance moment, visible wherever a founding constraint meets the capital required to keep building.
The synthesis observation
The AGI clause is the cleanest case study available of a contractual definition of AGI colliding with the capital structure built on top of it, and the capital structure winning. A provision written to wall AGI off from a single corporation became the price of that corporation’s continued partnership, and was renegotiated — across October 2025 and April 2026 — from a unilateral, deal-ending trigger into a jointly-verified, consequence-free checkpoint. Unilateral declaration became panel verification. Access termination became access through 2032. Payment escalation became payment decoupling. The form of the mission survived; its force was traded for the capital and consent OpenAI’s restructuring required.
There is no single answer. Anyone offering one is selling something. What is unambiguous is that the most important sentence in artificial intelligence for six years turned out to be a negotiable contract term — binding until it was inconvenient, protective until it was expensive, and definitional about a thing the parties never actually defined. The clause did not resolve what AGI is. It demonstrated something arguably more important: that when the definition of AGI stands between a company and the capital it needs, the definition is what moves.
That is the structural editorial question the clause sits on top of. It is governance encoded as contract, and therefore governance that is negotiable. It is a mission constraint defused at the table rather than enforced at the cliff. It is a hard question made moot rather than answered. And it is the clearest evidence yet of how AI’s founding ideals fare when they meet the balance sheet — not breached, not betrayed, but renegotiated into a form that no longer constrains the thing it was written to constrain.
About the Author
Thorsten Meyer is a Munich-based futurist, post-labor economist, and recipient of OpenAI’s 10 Billion Token Award. He spent two decades managing €1B+ portfolios in enterprise ICT before deciding that writing about the transition was more useful than managing quarterly slides through it. He runs StrongMocha News Group, a network of more than 450 niche WordPress magazines built on the DojoClaw editorial engine. More at ThorstenMeyerAI.com.
Related Reading · the AI Governance & Corporate Structure track
This dispatch
- This piece · The clause · how the Microsoft–OpenAI AGI provision was renegotiated from a deal-ending trigger into a consequence-free verification step, and what that reveals about governance ideals under capital pressure · synthesis-deep dominant, structural-slate and labor-rose balance
The track
- The cleaner cap table · AI Governance 02 · Anthropic’s PBC-from-inception structure and the Long-Term Benefit Trust — the structural mirror to OpenAI’s bolted-on governance · the natural companion to this piece
- The calendar technicality · AI Governance 01 · the Musk v. Altman verdict and the statute-of-limitations dismissal — the litigation backdrop to OpenAI’s restructuring
- Forthcoming · The S-1 disclosure burden · what OpenAI and Anthropic will have to disclose about governance and AGI risk when they file · empirical-clay register
- Forthcoming · The nonprofit-conversion playbook · how the IRS and state attorneys general shape AI-lab restructurings · structural-slate register
Adjacent tracks
- The CFO’s new operating system · Enterprise Reorg 01 · the commercial deployment of the models this clause governs
- The pyramid cracks · Enterprise Reorg 02 · the consulting compression driven by the agents OpenAI and Microsoft are commercializing
Sources
The original clause
- FlowHunt · The OpenAI–Microsoft AGI Clause Battle — the 2019 provision restricting Microsoft’s access once AGI is achieved · OpenAI’s board holding unilateral declaration authority · AGI as systems surpassing humans in most economically valuable tasks · the contested, undefined criteria · OpenAI needing Microsoft’s approval to restructure as a PBC · flowhunt.io
- The AGI Clause: A Ticking Time Bomb — the clause’s structure: Microsoft loses access if AGI is achieved, OpenAI declares it unilaterally, no agreed definition · “a time bomb without a timer” · Microsoft’s reported ~80% internal urgency to resolve · mission vs. monetization framing · goodreads.com
- Frontier Stocks — the $100B potential-profits threshold paired with the “surpass humans in most economically valuable work” language · Microsoft believing AGI far off, OpenAI believing it imminent · automatic termination of cooperation on declaration · frontierstocks.substack.com
The path to resolution
- European AI & Cloud Summit — the timeline: summer 2025 negotiations at “boiling point,” OpenAI considering antitrust regulators · Sept 9 2025 Microsoft integrates Claude into Office 365 Copilot · Sept 10 OpenAI’s $300B Oracle deal · Sept 11 tentative agreement (non-binding MOU) resolving AGI-clause, exclusivity, and restructuring tensions · cloudsummit.eu
The October 2025 recapitalization
- GeekWire · Microsoft gets 27% stake — Oct 28 2025 · 27% of OpenAI Group PBC (~$135B), down from 32.5% · incremental $250B Azure commitment · any AGI declaration now subject to independent expert-panel verification · geekwire.com
- Official Microsoft Blog · The next chapter — the primary source: 27% on as-converted diluted basis · revenue share remains until the expert panel verifies AGI, paid over a longer period · $250B incremental Azure · Microsoft loses right of first refusal on compute · blogs.microsoft.com
- Computing.co.uk · Microsoft takes 27% stake — the declaration power shifts from OpenAI’s board to a jointly-established independent panel operating with autonomy and transparency · Bret Taylor on the nonprofit retaining control with a path to resources · Microsoft’s compute limits and revenue-share if it develops AGI pre-verification using OpenAI tech · computing.co.uk
- DeepQuarry · Microsoft starts talking about OpenAI — the 8-K detail: IP rights for models and products through 2032 covering post-AGI systems with safety guardrails · research-method rights until AGI verification or 2030 · IP excludes consumer hardware · co-developed API products Azure-exclusive, non-API any cloud · deepquarry.substack.com
- Pure AI · Rewrite Partnership Rules — OpenAI can no longer unilaterally declare AGI; must be panel-verified · Foundation’s ~26% stake (~$130B), employees/investors 74% · performance warrant (10x in 15 years) · all Foundation directors on the for-profit board except Zico Kolter (non-voting observer) · pureai.com
- Yahoo/Reuters · 27% stake — Microsoft’s IP rights (excl. consumer hardware) through 2032, including post-AGI models · loses rights to OpenAI’s research at panel confirmation or 2030, whichever first · revenue-sharing ends when AGI confirmed · finance.yahoo.com
The April 2026 amendment
- MindStudio · OpenAI–Microsoft Deal Restructured — April 27 2026 amendment · the next-day AWS Bedrock launch · locked-in 20% revenue share, removed AGI-trigger from payments, non-exclusive license, immediate AWS launch · OpenAI’s inference demand exceeding any single cloud as the underlying driver · mindstudio.ai
- Nerd Level Tech · AGI Clause Dies, AWS Bedrock Live — the April 27 amendment leaves the October panel-verification regime in place but verification “no longer determines whether Microsoft’s license continues” · cloud exclusivity over, Azure keeps a first-shot clause · OpenAI Group ~$500B in the October round · nerdleveltech.com
- Cryptopolitan · OpenAI to save $97 billion — ~$97B in projected savings through 2030 from the capped revenue share and removed AGI-triggered payment clauses · AGI verification still ends research-IP rights but no longer triggers revenue escalation · products launch first on Azure “unless Microsoft cannot and chooses not to support” · cryptopolitan.com
Key reference figures crystallized
- The original clause (2019): AGI ends Microsoft’s access · OpenAI’s board declares AGI unilaterally · no agreed definition · “surpass humans in most economically valuable work” + ~$100B potential-profits marker · “a time bomb without a timer”
- The boiling point (2025): summer negotiations at boiling point · OpenAI weighs antitrust route · Microsoft ~80% internal urgency · Sept 9 Claude into Copilot · Sept 10 $300B Oracle · Sept 11 tentative MOU
- October 28, 2025: OpenAI Group PBC, nonprofit Foundation retains control (~26%, ~$130B) · Microsoft ~27% (~$135B), down from 32.5% · $250B incremental Azure · unilateral declaration → independent-panel verification · IP through 2032 incl. post-AGI · research rights until verification or 2030 · Microsoft loses compute right of first refusal · Group valued ~$500B
- April 27, 2026: cloud exclusivity ends · AWS Bedrock live next day · Azure first-shot clause retained · revenue share capped and decoupled from AGI · verification no longer determines license continuation · ~$97B OpenAI savings through 2030
- What survives: independent expert panel verifies any AGI declaration · verified AGI ends Microsoft’s research-IP rights (or 2030, whichever first) · AGI still operationally undefined
- The structural pattern: governance encoded as contract → negotiable · mutual-hostage clause → renegotiated not enforced · hard question made moot not answered · form of mission survives, force traded away
