They helped create OpenAI to keep transformative artificial intelligence from being concentrated in a few hands. A decade later, Elon Musk and Sam Altman were fighting over who had betrayed that promise—and who should control the institution built around it.
This is the history of the alliance, the rupture, the competing companies, the takeover offer and the lawsuits. It follows the documents rather than treating either billionaire’s version as the whole story.
Where the conflict stands
The crucial distinction
Musk lost the main trial in May 2026 because his claims were brought too late. The outcome was a victory for OpenAI, Altman, Brockman and Microsoft in that proceeding. It was not a verdict establishing that every accusation of mission betrayal was false, or that every business decision was justified. [46]
There is more than one legal fight. The charitable-mission case, a trade-secrets case, and the Apple–OpenAI antitrust case have different parties, claims and procedural histories. A headline saying “Musk loses to OpenAI” cannot safely be applied to all three. The separate trade-secrets appeal was on a briefing schedule in September; a 14 September motion sought to dismiss Apple from the competition case while expressly continuing the claims against OpenAI. [53] [50]
Beyond court, the rivalry now reaches the software that people use. OpenAI announced that Cursor’s access to its models would end on 12 November 2026 following Cursor’s acquisition by SpaceX. That was a future deadline at this report’s research cutoff, not a shutdown that had already occurred. [55] [56]
Scope and method. This is a detailed account of the documented public dispute, not a claim to possess every private conversation or sealed filing. “Alleged” identifies a party’s claim; testimony is attributed to its witness; a ruling is described by what it actually decided. Analysis is identified separately. No inference about private motive is presented as a proven fact.
What they are fighting over
The personal feud is the easiest part to watch. The institutional question is harder: when donors support a nonprofit research organisation, how much of its research, workforce and future commercial value can subsequently sit in a business with private investors? There is a further question of power: who can enforce the mission when the chief executive, employees, investors and the nonprofit board disagree?
Musk’s case framed the transformation as a betrayal of commitments that induced him to provide money, credibility and support. OpenAI’s response framed him as a former insider who had himself contemplated a commercial structure, wanted control, and later used litigation against a successful competitor. Both are interested accounts. The emails matter because they complicate each simple version. [26] [05]
Our analysis: supporting some form of commercial subsidiary is not necessarily consent to every later transaction. Equally, a person who considered owning or controlling a profit-making successor cannot convincingly be described as having opposed commercialisation in every circumstance. Those propositions can both be true. Neither, alone, determines whether a particular promise was enforceable or a particular restructuring lawful.
| Actor | Place in the dispute |
|---|---|
| Elon Musk | Founding co-chair and donor; left OpenAI’s board in 2018; later founded competitor xAI and became a plaintiff. His donations were not a shareholding in the nonprofit. |
| Sam Altman | Founding co-chair and later chief executive. His managerial influence must be distinguished from the legal powers of OpenAI’s board. |
| Greg Brockman and Ilya Sutskever | Core co-founders and early leaders. Their correspondence helps explain why the story cannot be reduced to a two-person partnership. |
| OpenAI nonprofit / Foundation | The charitable institution at the centre of the mission dispute; retained control under the completed 2025 structure. |
| OpenAI’s commercial entities | The businesses that raise investment, employ resources and sell products. Their structure changed over time. |
| Microsoft | Investor, cloud partner and licensee; also a defendant in the main case. Investment and licensing are not the same as ownership of the charity. |
| xAI, X and SpaceX | Musk-linked businesses whose roles evolved across the rivalry. Their claims and transactions should not automatically be attributed to Musk personally. |
Founding and board roles: [01] [02] Structure and partnership: [39] [42]
2015–2018: the alliance and the rupture
The original disagreement about concentration of AI power developed into a disagreement about control of OpenAI itself.
OpenAI is introduced as a nonprofit
The launch presented a research organisation pursuing broad public benefit without the pressure to generate financial returns. Musk and Altman were co-chairs; Brockman and Sutskever held central technical roles. The announcement listed a group of backers committing a collective $1 billion. That was a funding commitment by multiple supporters—not evidence that Musk personally handed over $1 billion. [01]
The cost of competing changes the conversation
Correspondence published by OpenAI describes discussions about raising much more money and adopting a commercial structure. Profit was being discussed before Musk left. The source is a litigant’s selected release, not a neutral account. [06]
A struggle over the proposed company
The released material records disputes over Musk’s desired initial control, ownership and leadership, alongside a proposed public-benefit company. Other founders resisted concentrating authority in one person. Control was already a central disagreement. [06]
Reassurances become future evidence
Musk later relied on assurances about continuing the nonprofit mission. In opposing summary judgment, his lawyers argued that these commitments induced continued support and could not be erased by general discussions about fundraising. That filing lays out the plaintiff’s theory; it is not itself proof that a fraudulent promise was made. [09]
The Tesla route is discussed
OpenAI’s published correspondence describes a proposal to bring the AI effort into Tesla. The practical argument concerned the resources necessary to compete. The governance implication was substantial: an organisation conceived as an independent counterweight could instead become connected to one founder’s existing commercial enterprise. Whether that was a sensible funding solution and whether it honoured the original purpose are different questions. [06]
Musk leaves the board
The public announcement on 20 February cited a potential conflict as Tesla became more focused on AI, and said Musk would continue to donate and advise. Later accounts describe deeper disagreements over control and direction. Readers should preserve both layers: the contemporaneous public explanation and the later documentary dispute, rather than silently replacing the former with the latter. [02] [05]
The Charter states the mission more explicitly
The Charter describes broadly distributed benefits, avoidance of undue concentration of power, long-term safety and cooperation. It also anticipates that safety could eventually reduce traditional publication. This matters because “open” can refer to shared benefit, published research, model access or downloadable code; those are not identical commitments. [03]
The scale of the funding problem remains
OpenAI’s March 2024 response reproduced correspondence in which Musk discussed the need for billions per year to compete. It supports the narrower conclusion that he recognised very large capital requirements. It does not settle whether the later Microsoft arrangements or changes to investor returns complied with earlier obligations. [05]
2019–2023: commercial success and a new rival
After Musk’s departure, OpenAI’s funding model and public profile changed. By 2023 he was criticising the organisation while building a competitor.
The capped-profit OpenAI LP is announced
OpenAI created a commercial investment structure under nonprofit control. The first investors’ returns were capped at 100 times their investment, with different terms possible for later rounds; value above the caps was intended for the nonprofit. This was not an ordinary charity receiving donations, but neither was the announced arrangement a simple sale of the charity to investors. [04]
Microsoft commits $1 billion
The partnership paired investment with Azure computing infrastructure and plans to commercialise technology. It gave OpenAI access to capital and computing on a scale donations alone had not supplied. It also made a major technology company central to an organisation originally presented as a counterweight to concentrated AI power. That tension exists regardless of one’s view of Musk. [10]
Microsoft receives an exclusive GPT-3 licence
Microsoft announced an exclusive licence to the model technology. OpenAI separately said its API would continue to provide access to users and developers. An exclusive technology licence, general access through a hosted API and nonprofit governance are separate layers; describing all three as “Microsoft owns OpenAI” collapses distinctions essential to understanding the dispute. [11] [12]
ChatGPT becomes a public-facing product
OpenAI released ChatGPT as a conversational research preview and invited feedback. The importance to this history is the transition in visibility: the institution was now associated with a product ordinary people could use, rather than only papers, demonstrations and developer services. The relationship with Musk would increasingly be argued in front of that wider audience. [13]
The Microsoft relationship expands again
Microsoft announced a multiyear, multibillion-dollar partnership. A later UK competition-authority summary described total investment exceeding $13 billion, including $10 billion in January 2023. These are historical investment figures, not a present-day valuation and not the amount Musk donated. [14] [15]
GPT-4 raises the stakes
OpenAI announced GPT-4. The model later became part of Musk’s legal argument about how advanced the technology had become and who should benefit from it. A product launch and a litigant’s assertion that a system qualifies as artificial general intelligence are different things; the complaint did not turn that assertion into an established judicial finding. [16] [26]
Musk signs a call for a six-month pause
The Future of Life Institute letter urged a pause in training systems more powerful than GPT-4, with safety protocols and governance work. Its scope matters. It was not a request to stop every AI experiment, every chatbot or all commercial research. Any accusation that a signatory violated the pledge must address the actual proposed threshold. [17] [18]
xAI is publicly launched
Musk introduced an AI company to compete in the same field. The public launch made his role more complicated: he was both a critic of OpenAI’s mission and the leader of a rival business. Commercial incentives do not automatically invalidate a criticism, but they are material context when assessing litigation, recruitment disputes and demands affecting a competitor. [19]
Grok offers a competing identity
xAI introduced Grok with an emphasis on humour, a rebellious personality and access to information through X. That was product positioning as well as an implicit criticism of other chatbots. Claims that one system is more truthful, less ideological or safer require evidence about its behaviour; a founder’s preferred branding cannot establish them. [21]
OpenAI’s board fires Altman
The board said Altman had not been consistently candid with it and that this impaired its oversight. The significance extends beyond office politics: the nonprofit board was supposed to be the mechanism protecting the mission. Its decision exposed a serious breakdown between formal oversight and executive leadership. The announcement gave the board’s account, not a court-tested determination of misconduct. [23]
Altman returns; the board changes
OpenAI’s 29 November announcement confirmed Altman’s return as chief executive and an initial board of Bret Taylor, Larry Summers and Adam D’Angelo. The crisis demonstrated that possessing formal removal powers does not guarantee an oversight body can exercise them without destabilising the institution. That is an analytical lesson, not proof that every objection raised by the former board was right. [24]
2024: the argument becomes litigation
The dispute moved through a first lawsuit, withdrawal, a new federal action and competing releases of historical documents.
Musk files the first lawsuit
The San Francisco complaint alleged that OpenAI and its leaders had departed from a founding agreement centred on nonprofit public benefit and openness, particularly through the Microsoft relationship. It invoked GPT-4 and advanced-AI obligations. The pleading states Musk’s allegations; it did not establish a binding founding contract, prove fraud or legally determine that GPT-4 was AGI. [26]
OpenAI answers with emails
OpenAI denied the central narrative and published correspondence about funding, control and the founders’ discussions. Its account said Musk contributed less than $45 million and that other donors supplied more than $90 million. These were the company’s stated figures, not a newly adjudicated accounting. The release also made clear why selected emails would become a major weapon in the public dispute. [05]
The company publishes its board-crisis review summary
OpenAI said the WilmerHale review identified a breakdown in trust and concluded Altman’s conduct did not require removal. Its summary said the firing was not driven by product-safety, security, financial or development-pace concerns. This was a company-published summary of a commissioned review, not the full underlying evidentiary record or an independent court judgment. [25]
xAI releases Grok-1 weights
xAI made Grok-1’s base-model weights and architecture available under Apache 2.0. It was a meaningful release that others could inspect and build on. It was not disclosure of all training data, every later Grok model or the entire production service. Calling it either “nothing” or “everything is open” would misdescribe the release. [22]
Altman discusses the broken relationship
Speaking to Lex Fridman, Altman called the litigation unbecoming of a builder while saying he still respected Musk and hoped for an amicable relationship. He also defended low-cost public access as one form of openness and discouraged other founders from starting a nonprofit only to add a commercial arm later. These statements show Altman’s account, not Musk’s private motives. [57]
Apple announces ChatGPT integration
Apple announced an arrangement to make ChatGPT available through parts of its device experience, with user choice and permission mechanisms described in the announcement. This introduced another major distribution platform into the competition. It should not be confused with Microsoft’s investment and cloud agreements, or with the later antitrust allegations about Apple’s treatment of competing apps. [48]
Musk withdraws the state-court case
Musk dropped the original action before a scheduled hearing. The withdrawal did not produce a merits ruling vindicating either side. It was a procedural end to that particular complaint, not the end of the dispute. [27]
A federal complaint restarts the fight
The new federal action in California advanced a broader account of deception and the commercial transformation. It included fraud and other theories, rather than merely reproducing the earlier state case unchanged. A complaint can be forceful and detailed while still containing allegations that must survive legal challenges and be proved. [28]
The federal case broadens
Amended pleadings brought xAI and Shivon Zilis into the plaintiff side and extended the dispute to Microsoft and competition issues. The March 2025 injunction ruling describes that expanded procedural setting. From this point, “Musk versus Altman” was an especially incomplete shorthand for a case involving several people, entities and distinct causes of action. [29]
More early correspondence is released
OpenAI published a longer documentary account of the founding-era negotiations. It argued that Musk’s objection concerned losing control rather than profit itself. That interpretation deserves scrutiny alongside the documents used to support it. [06]
OpenAI proposes another structural change
OpenAI outlined a public-benefit corporation that would control commercial operations, while the nonprofit would receive equity valued by advisers and pursue charitable initiatives. This proposal is essential historical context because it differed from the retained-nonprofit-control arrangement announced the following May. Accounts of the dispute should not treat an abandoned proposal as the final corporate structure. [37]
2025: the bid, the counterattack and the restructure
Money, litigation and governance converged. The $97.4 billion offer was only one part of a much larger argument over charitable assets.
Competition concerns extend beyond the feud
The FTC published a staff report about major cloud providers’ relationships with AI developers, including Microsoft–OpenAI. It raised concerns about matters such as access to computing, switching costs and sensitive information. The report shows that structural competition questions are legitimate subjects of investigation. It was not a ruling that Musk had proved his case or that a named partnership was unlawful. [41]
Stargate triggers a public funding argument
OpenAI announced a project intending to invest $500 billion in US AI infrastructure over four years, beginning with $100 billion. Musk questioned whether the backers had the money; Altman rejected the criticism. The announced total was an investment ambition, not proof that $500 billion had already been spent or was sitting in an account. [32] [33]
A Musk-led consortium offers $97.4 billion
The offer targeted the nonprofit’s assets. It was not simply a stock-market purchase of all of OpenAI for that price. Altman characterised it as an attempt to interfere with the company. The bid also created a valuation question: what would count as fair compensation for charitable assets during a restructuring? A proposed price is evidence of an offer, not an independently established fair value. [34]
Musk ties withdrawal to stopping the conversion
Musk’s side said the offer would be withdrawn if OpenAI halted the contested conversion and preserved the charity’s mission. That conditional position connected the acquisition effort directly to the governance dispute. It still did not establish whether the bid was principally a genuine acquisition strategy, litigation pressure or a mixture of both. [35]
The board rejects the offer
OpenAI’s board rejected the bid and said the organisation was not for sale. Rejection did not erase the underlying questions about control and asset valuation, but neither did a bidder acquire an entitlement to the charity by offering a large number. [36]
The preliminary injunction is denied
Judge Yvonne Gonzalez Rogers declined the requested interim relief. A preliminary injunction concerns whether demanding legal requirements for immediate intervention are met; denial does not necessarily resolve every claim after a full trial. Conversely, a court allowing a case to continue does not certify the allegations as true. [29]
OpenAI files counterclaims
OpenAI accused Musk of using litigation and related conduct to interfere with its business, including characterising the bid as part of a campaign against the company. Those are counter-allegations, not findings merely because they appear in a court filing. This report does not independently establish a final disposition of every counterclaim. [30]
The court narrows the federal case
A partial-dismissal order removed claims while allowing others to continue. It is an important checkpoint between filing and trial: the sprawling original accusations should not be described as though every legal theory reached a jury unchanged. The order itself is the best source for the claim-by-claim result. [31]
OpenAI says nonprofit control will remain
The revised plan retained the nonprofit’s control of the commercial company, while changing the commercial entity into a public-benefit corporation with conventional stock. OpenAI referred to discussions with civic leaders and state attorneys general. Musk’s litigation formed part of the surrounding pressure, but the public record does not justify attributing the change solely to him. [38]
X and xAI sue Apple and OpenAI
A separate Texas antitrust complaint alleged exclusionary conduct involving Apple’s platform and its relationship with OpenAI. These were competition allegations by rival businesses, not another version of the donor’s charitable-trust claim. The complaint did not establish that app rankings or integration arrangements were unlawful. [49]
A separate trade-secrets battle begins
xAI accused OpenAI of benefiting from the recruitment of employees and confidential information. Employee movement and unlawful acquisition of trade secrets are distinct issues: a company can lawfully hire a rival’s staff without being entitled to the rival’s protected information. The sufficiency of the alleged connection became central to the dismissal rulings. [52]
The recapitalization is completed
OpenAI described a nonprofit Foundation controlling OpenAI Group, a public-benefit corporation. At closing, it reported a 26% Foundation stake, roughly 27% for Microsoft and 47% for employees and other investors. The Foundation’s special governance rights included appointing and replacing the Group’s directors. California’s attorney general separately issued a statement about the recapitalization and safeguards. [39] [40]
2026: trial, defeat and a widening commercial conflict
The main trial ended on timing. Other proceedings and business decisions continued to reshape the rivalry.
The central case survives summary judgment
The court found disputes requiring trial rather than deciding the central historical questions in OpenAI’s favour on the written record. Its opinion discussed donor reliance, knowledge and communications. Survival at this stage meant the evidence needed testing; it was not a finding that Musk had already proved deception or a breach of charitable duties. [08]
Brockman’s notes become a public battleground
OpenAI responded to Musk’s use of private notes suggesting ethical discomfort with a conversion without him. The company argued that the passages concerned a hypothetical route the founders rejected. This is precisely where short excerpts can mislead: the disagreement concerns context and what actually followed, as well as the words themselves. [07]
SpaceX acquires xAI
SpaceX’s later prospectus records the acquisition. It adds another corporate layer to the rivalry: decisions involving Musk’s AI operations can now intersect with the resources and businesses of SpaceX. Corporate consolidation is relevant to the public question of concentrated AI power, regardless of which founder is criticising the other. [54]
Microsoft announces less exclusive terms
Microsoft said its OpenAI intellectual-property licence would become non-exclusive and continue through 2032. It remained the primary cloud partner, but OpenAI could serve products through other cloud providers. Descriptions based solely on the 2019 or 2023 arrangements therefore became historically incomplete. The revised deal still left Microsoft a major shareholder and partner. [42]
The main trial opens
The Oakland proceeding moved from jury selection into opening arguments and testimony, with Musk taking the stand. His case sought very substantial monetary and structural remedies. Claims in pleadings, evidence admitted during trial and the final basis of decision must be kept separate; the most dramatic headline from a witness does not determine the outcome. [43]
Brockman is questioned about money and mission
Court reporting described questions about Brockman’s private notes, potential personal wealth, investments and enormous equity interest. He defended his commitment to the mission and the value created by the company. This evidence gave substance to questions about incentives; it did not produce a final finding that personal enrichment had caused an unlawful diversion of charity assets. [44]
Former insiders challenge Altman’s leadership
Former directors Helen Toner and Tasha McCauley gave recorded testimony about oversight and candour concerns. Former employee Rosie Campbell described the treatment of safety work. These accounts deserve inclusion alongside the company’s favourable review summary. They are attributed testimony, contested in an adversarial proceeding—not a judicial finding that all of OpenAI’s safety claims were false. [45] [25]
Musk loses on the statute of limitations
The advisory jury found the claims untimely, and Judge Gonzalez Rogers adopted that result and dismissed Musk’s claims. The decisive issue was when he knew or should have known enough to sue. The jury did not proceed to settle the whole underlying mission-betrayal narrative. Musk’s side announced that it planned to appeal. [47]
The trade-secrets case is dismissed again
Judge Rita Lin dismissed xAI’s revised claims against OpenAI without permission for another amendment. Reporting describes the court finding the pleaded theory speculative and insufficient to connect OpenAI to the alleged misconduct. This is a distinct case with a distinct judge and legal basis from the May charitable-trust trial. [51]
Fees are sought; a separate appeal is opened
OpenAI sought more than $1 million in legal costs in the trade-secrets dispute; a request is not an award. The Ninth Circuit opened appeal 26-4486 on 15 July. Its docket identifies the originating case as 3:25-cv-08133-RFL, preventing confusion with Musk’s principal mission lawsuit. [52] [53]
Cursor joins SpaceX
Cursor announced that it was joining SpaceX. An AI coding product that uses outside model providers was now part of a business group competing with one of those providers. This made the feud relevant to developers’ model choices and supplier relationships, not merely the reputations of the founders. [55]
OpenAI announces a future Cursor cutoff
OpenAI said it would terminate Cursor’s access to its models on 12 November. It justified the decision by reference to competition and alleged prior contractual violations involving Musk’s companies. That justification is OpenAI’s account, not an independent finding here. The announcement set a future transition date; it did not mean access stopped in August. [56]
The trade-secrets appeal has a briefing schedule
The public appellate docket records an extension setting OpenAI’s answering brief deadline for 23 October 2026, with a later optional reply. This confirms a live appellate process in that separate matter. It does not establish that the appeal will succeed, nor does it tell us the status of an appeal in the main charitable-mission case. [53]
Apple’s exit is sought; OpenAI remains targeted
An unopposed motion stated that the claims against Apple had been resolved and asked for dismissal with prejudice. It expressly said the claims against OpenAI would continue. The motion did not disclose settlement terms, and it should not be described as proof that Apple paid money, admitted fault or that the court had already entered the requested order. [50]
Follow the money—and the control
The feud produces very large numbers that refer to different things. Confusing a donation with an investment, an investment with a valuation, or a bid with a court award makes the story more dramatic and less accurate.
| Figure | Meaning and limitation |
|---|---|
| $1 billion, 2015 | A collective founding funding commitment, not Musk’s personal cash donation. [01] |
| About $38 million | Charitable donations described in the January 2026 opinion. Different accounts use different totals and categories; this should not be silently substituted for every measure of Musk’s support. [08] |
| More than $13 billion | Microsoft investment described by the CMA in 2025. This does not mean it owned the nonprofit. [15] |
| $97.4 billion | The consortium’s 2025 proposed acquisition price for nonprofit assets. The offer was rejected. It was not a completed payment or court valuation. [34] [36] |
| Up to $500 billion | The four-year Stargate investment ambition announced in January 2025, not a statement of completed expenditure. [32] |
| 26% / 27% / 47% | The Foundation, Microsoft and remaining holder stakes reported at the October 2025 recapitalization closing. Historical closing percentages are not a guarantee of an unchanged September 2026 capitalization table. [39] |
Donation is not ownership
A donor may have rights relating to promises or restrictions without owning the charity in the way a shareholder owns stock. Musk’s early importance therefore does not, by itself, establish a right to appoint himself chief executive, absorb the organisation into Tesla or veto every transaction. Conversely, calling his contribution a donation does not by itself dispose of his argument about restricted charitable purposes. Those were matters to be examined through the promises, documents and applicable law. [09]
Nonprofit control and private profit can coexist
The 2019 model deliberately combined nonprofit governance with capped investor returns. By 2025 the commercial structure had changed again. The essential analytical distinction is between economic participation—who benefits financially—and governance rights—who can appoint directors and enforce the mission. A minority economic stake can carry special control rights. A very large investment can carry commercial influence without ownership of the charity. [04] [39]
Our analysis: formal control is only the start of an accountability assessment. Readers should also ask who funds the next computing expansion, whether directors can withstand employee and investor pressure, how conflicts of interest are handled, and what happens when safety constraints reduce revenue. The 2023 board crisis makes those questions practical rather than hypothetical. [23] [24]
The competing claims, tested against the record
“Musk always opposed making money from OpenAI.”
The record is more complicated. Published correspondence documents commercial and control proposals involving Musk. His stronger argument is not that no commercial idea was ever discussed, but that later actions violated the particular commitments on which he relied. OpenAI’s stronger response is not merely that fundraising was necessary, but that his own proposals undercut his public description of the original bargain. Neither proposition eliminates the need to inspect the actual terms. [06] [09]
“Altman simply stole a charity.”
That is an accusation, not this report’s finding. Musk alleged wrongful diversion and broken charitable commitments. OpenAI argued that commercial funding advanced the mission, and ultimately retained nonprofit control in the completed restructuring. Retaining control does not automatically answer every historical complaint, but it must appear in any honest description of the result. [28] [38] [40]
“The internal notes prove the entire case.”
They raise serious questions but require context. Musk’s side used Brockman’s writing to contrast private intentions with assurances to a donor. OpenAI said the passages concerned hypothetical courses of action that were not followed. Trial questioning also explored Brockman’s financial incentives. Extracts can be probative without proving every step of a legal claim, including reliance, causation and timeliness. [07] [44]
“OpenAI won, so nothing improper happened.”
The actual outcome is narrower. The main trial ended on a filing deadline. It was a real legal defeat for Musk and a real litigation victory for the defendants, but it left the larger factual and ethical dispute without a complete merits verdict. Treating timing as an irrelevant technicality is misleading; treating it as universal factual exoneration is misleading too. [47]
“Musk’s competitor status makes every concern invalid.”
It creates a conflict of incentives, not an automatic rebuttal. A rival can identify a genuine governance problem while also benefiting from restrictions on a competitor. The proper response is to examine the evidence and proposed remedy. Would the remedy protect the charitable mission, transfer advantage to xAI, concentrate power elsewhere, or do some combination of those things? [19] [49]
“Microsoft owns or controls all of OpenAI.”
That is too crude. Microsoft’s investment, model licences, cloud role and reported equity are significant. They must still be distinguished from the Foundation’s legal governance rights. The April 2026 agreement also made model and product IP licensing non-exclusive. An accurate critique of dependency should use the actual arrangement at the relevant date. [42]
The lawsuit tracker
The status below is limited to the latest documents and reporting checked for this edition. An announced intention to appeal is not interchangeable with a docketed appeal. A request for fees is not an award, and a motion to dismiss a party is not itself the resulting order.
1. The original California state action
Filed February 2024 · withdrawn June 2024
Musk’s initial suit against OpenAI and its leaders alleged departure from the founding bargain. It ended by withdrawal, without a merits judgment. The later federal action was a new proceeding, so its rulings should not be retroactively described as decisions in this withdrawn case. [27] [26]
2. The principal federal mission case
Northern District of California · 4:24-cv-04722-YGR
The case went through amended pleadings, an unsuccessful preliminary-injunction request, partial dismissals and summary-judgment proceedings before trial. The remaining charitable-trust and unjust-enrichment dispute ended in May 2026 on timeliness. Microsoft’s related liability was also defeated. Musk’s lawyers announced plans to appeal; this report has not independently verified a later appellate ruling in this main case. [29] [31] [08] [47]
OpenAI also filed counterclaims. The May result should not be used to invent a separate damages award against Musk or a final determination of every counter-allegation. [30]
3. xAI’s trade-secrets case
District case 3:25-cv-08133-RFL · Ninth Circuit appeal 26-4486
The claims against OpenAI were dismissed after amendment. The public appeal docket shows an opening brief filed in August and an answering-brief deadline extended to October 2026. That is verified appellate activity in this case, not evidence of the procedural status of the main donor litigation. OpenAI’s legal-fee request is separately reported; no fee award is asserted here. [51] [52] [53]
4. X/xAI’s Apple–OpenAI competition case
Northern District of Texas · 4:25-cv-00914-P
The original complaint alleged anticompetitive conduct. By 14 September 2026, the plaintiffs sought to dismiss Apple with prejudice, stating those claims were resolved, while continuing against OpenAI. The public motion did not supply the terms of that resolution. “Resolved” does not tell the reader who conceded what, whether money changed hands, or whether any allegation was admitted. [49] [50]
Safety, openness and the credibility problem
What does “open” actually mean?
There are several different tests: can a researcher read the paper, inspect the code, download the model weights, reproduce the training, audit the data, or access the product through a paid service? Passing one does not mean passing them all. The founding announcement favoured sharing research; the later Charter anticipated circumstances in which safety could limit publication. A careful critique should identify the particular promise and the particular departure. [01] [03]
The same standard applies to xAI. Its Grok-1 release supplied weights and architecture for a base model. That deserves recognition without turning it into a claim that every later system or all training material is publicly inspectable. Openness is a property of a particular release and licence, not a permanent moral status conferred on a company. [22]
Did launching xAI contradict the pause letter?
The juxtaposition is politically potent: Musk signed a call for restraint and also established a rival. But the letter’s request was specifically about training systems beyond GPT-4. xAI’s framework dates the company’s founding to March 2023 and Grok-1 training to August. Those facts alone do not demonstrate that its activity crossed the letter’s proposed threshold. The sound criticism concerns incentives and transparency, not an unsupported declaration that any AI work necessarily broke the request. [18] [20]
What did the testimony add?
Court reporting recorded former insiders’ criticisms of oversight, candid communication and the standing of safety work. Their accounts make it inadequate to answer every criticism with a mission statement. OpenAI’s response and its commissioned-review summary must also be read. The existence of adverse testimony is not equivalent to a verdict that a product was unsafe or that a witness’s full account was accepted. [45] [25]
Our analysis: both leaders benefit from presenting their preferred organisation as the responsible home for powerful AI. The useful questions concern observable safeguards: disclosure of serious failures, independence of safety decisions, meaningful outside evaluation, enforceable limits, and whether leadership accepts constraints when they are inconvenient. A clash of personalities cannot substitute for those tests.
What the feud reveals about power
The following is editorial analysis. The strongest criticism of Musk is the mismatch between a campaign against concentrated power and his own documented interest in decisive control. The strongest criticism of OpenAI’s leadership is that public-benefit language can coexist with arrangements creating enormous private incentives—and that reassuring language does not remove the need for independent governance. These criticisms are compatible. Readers do not have to choose a billionaire in order to ask who can hold the institution accountable.
The money creates a genuine dilemma rather than a ready-made excuse. Advanced AI requires expensive infrastructure, and a laboratory that cannot fund its work may fail to achieve its mission. But “we need capital” is not a complete answer to questions about who receives economic rights, what donors were promised and whether commercial dependence compromises oversight. Conversely, a large early donation is not a mandate for the donor’s personal rule.
The record also shows why remedies matter as much as accusations. Stopping a transaction, changing directors, preserving nonprofit control, opening research, transferring assets and paying damages are different interventions. One might improve accountability while another simply rearranges which wealthy actor benefits. A report on this dispute should examine the institutional consequences of each remedy rather than treating every attack on Altman as inherently public-spirited or every defence of OpenAI as inherently mission-preserving.
The litigation has produced information the public would otherwise have struggled to obtain: negotiating positions, internal doubts, financial incentives and accounts from former insiders. Its value as a source of evidence survives the failure of a claim. At the same time, litigation is an adversarial selection process. A party’s best exhibit is not necessarily the most representative document, and a vivid quotation does not explain the entire surrounding relationship.
The later competition disputes make the practical stakes especially clear. Model access, distribution through major platforms, recruitment and computing partnerships shape who can build products and on what terms. These are issues for developers, employees and users as well as the founders. The Cursor announcement is one concrete example of a corporate rivalry reaching a customer-facing product. [56]
What remains unresolved
- The complete merits of the founding dispute. The principal trial’s time-bar outcome did not answer every contested historical or ethical question.
- The separate appellate and competition proceedings. The trade-secrets appeal and continuing OpenAI antitrust claims require their own updates; their outcomes cannot be inferred from the May verdict.
- The practical durability of nonprofit oversight. Written control rights matter, but future exercises of those rights will show how resilient the arrangement is under commercial pressure.
- The terms behind public announcements. The Apple resolution motion and Cursor access notice do not disclose every relevant private contractual detail.
- The effect on users and developers. The announced November 2026 model-access deadline lay ahead of this edition. This report does not describe its implementation or consequences as already known.
The outstanding procedural points are grounded in the documents linked above. The governance questions are questions for continuing scrutiny, not predictions of wrongdoing.
Sources and document register
The register distinguishes court rulings, court filings, company statements and independent reporting. A corporate announcement is strong evidence of what the company announced; it is not independent verification of every justification offered. A complaint proves that an allegation was made, not that it is true. Links to PDFs may open the original filing in your browser.
- OpenAI’s founding announcement11 December 2015 · contemporaneous company announcement
- OpenAI’s board-departure announcement20 February 2018 · contemporaneous company announcement
- The OpenAI Charter2018 · mission and governance commitments
- OpenAI LP: the capped-profit structure11 March 2019 · company announcement
- OpenAI’s response to Musk and selected emails5 March 2024 · defendant’s account and published correspondence
- The 2015–2018 correspondence released by OpenAI13 December 2024 · party-selected documents and interpretation
- OpenAI’s response about Brockman’s notes16 January 2026 · defendant’s account
- January 2026 summary-judgment opinion15 January 2026 · judicial opinion, reproduced by FindLaw
- Musk’s opposition to summary judgment6 January 2026 · plaintiff’s legal argument, PDF
- Microsoft’s first major OpenAI investment22 July 2019 · company announcement
- Microsoft’s GPT-3 licensing announcement22 September 2020 · company announcement
- OpenAI’s explanation of the GPT-3 licence22 September 2020 · company announcement
- The original ChatGPT launch30 November 2022 · product announcement
- Microsoft’s expanded OpenAI partnership23 January 2023 · company announcement
- CMA summary of the Microsoft–OpenAI partnershipMarch 2025 · UK competition authority, PDF
- GPT-4 announcement14 March 2023 · product announcement
- The six-month AI-pause open letterMarch 2023 · original campaign letter and signatories
- What the proposed AI pause covered31 March 2023 · campaign’s explanation
- Axios on xAI’s public launch12 July 2023 · independent reporting
- xAI’s Frontier Artificial Intelligence Framework31 December 2025 · company policy and history, PDF
- Introducing GrokNovember 2023 · xAI product announcement
- Grok-1 open-weight release17 March 2024 · xAI release announcement
- OpenAI’s announcement removing Altman17 November 2023 · board’s contemporaneous explanation
- Altman’s return and the replacement board29 November 2023 · company announcement
- OpenAI’s summary of the WilmerHale review8 March 2024 · company-published review summary
- Musk’s original San Francisco complaint29 February 2024 · allegations in a civil complaint, PDF
- AP on withdrawal of the first lawsuitJune 2024 · independent reporting
- Musk’s original federal complaint5 August 2024 · complaint, not a judicial finding
- Order denying a preliminary injunction4 March 2025 · judicial ruling, PDF
- OpenAI’s counterclaims against Musk9 April 2025 · defendant’s allegations, PDF
- May 2025 partial-dismissal order1 May 2025 · judicial ruling, PDF
- The Stargate announcement21 January 2025 · investment plans announced by OpenAI
- AP on the Stargate funding argument22 January 2025 · independent reporting
- Reuters on the $97.4 billion offerFebruary 2025 · Reuters reporting republished by Investing.com
- AP on Musk’s condition for withdrawing the bidFebruary 2025 · independent reporting
- AP on the board’s rejection of the bid14 February 2025 · independent reporting
- OpenAI’s December 2024 restructuring proposal27 December 2024 · proposed structure, not the final arrangement
- OpenAI’s decision to retain nonprofit control5 May 2025 · revised company proposal
- OpenAI’s completed recapitalization structure28 October 2025 · company description; percentages at closing
- California attorney general on recapitalization28 October 2025 · regulator statement
- FTC staff report on cloud–AI partnerships17 January 2025 · regulator research, not a liability ruling
- Microsoft’s April 2026 revised agreement27 April 2026 · company announcement
- AP on the opening of Musk’s trialApril 2026 · independent court reporting
- Brockman’s testimony and financial interestsMay 2026 · Courthouse News trial reporting
- Safety and candour testimony at trialMay 2026 · Courthouse News trial reporting
- AP on the 18 May 2026 verdict18 May 2026 · independent court reporting
- Courthouse News on the time-bar ruling18 May 2026 · independent court reporting
- Apple’s ChatGPT integration announcement10 June 2024 · company announcement
- X and xAI’s Apple–OpenAI antitrust complaint25 August 2025 · allegations in a separate case, PDF
- Motion to dismiss Apple while continuing against OpenAI14 September 2026 · unopposed motion, not itself a court order, PDF
- Dismissal of xAI’s trade-secrets claimsJune 2026 · independent court reporting
- OpenAI’s legal-fee request after dismissal14 July 2026 · independent reporting
- The separate trade-secrets appeal docketNinth Circuit 26-4486 · public docket mirror, retrieved through 2 September 2026
- SpaceX’s June 2026 prospectus5 June 2026 · corporate prospectus, PDF
- Cursor announces it is joining SpaceX14 August 2026 · company announcement
- OpenAI’s notice about Cursor access28 August 2026 · OpenAI’s account and announced future cutoff
- Altman’s March 2024 interview transcriptMarch 2024 · direct interview, the speaker’s account
Prepared for The Musk Reports. Research cutoff: 19 September 2026. This report is based on the linked public record. It does not claim access to sealed evidence or original interviews with the parties. Historical figures are dated to their source, and planned future events are labelled as such.
For the publication’s approach, read how this site sources its claims. Factual challenges and responses should follow the corrections and right-of-reply policy.