OpenAI Cuts Off Cursor After SpaceX Acquisition — Model Access as a Weapon
At 9:47 this morning, OpenAI posted an announcement on X. Seven hours, 2.01 million views, 6,698 likes.
In one sentence: Cursor has been acquired by SpaceX, and OpenAI has decided to cut off supply. The proposed cutoff date: November 12, 2026.
The announcement was written politely, saying it "respects the Cursor team," cares about the developer experience, and is willing to "do everything possible" to help everyone transition.
But when you take the announcement apart, every word is saying something else: model supply cutoff, once again publicly used as a weapon — but never this blatantly.
One announcement, three killer moves
Let's lay out the facts first.
OpenAI today formally notified SpaceX that it will terminate the contract to supply OpenAI models to Cursor. The official statement gave three reasons and arrangements, each more calculated than the last:
First, the cutoff date is November 12, and this is the maximum notice period allowed by the contract. OpenAI specifically emphasized "to maximize the time developers retain access." Translation: I'm following the rules, you can't fault me legally, and I stand on solid moral ground.
Second, the reason is stated very bluntly, uncommonly blunt. The official wording: based on the "experience" of Musk's companies violating contracts, we cannot be confident that SpaceX will use our technology in accordance with the terms of service. Note, this is not "business adjustment," not "strategic refocusing" — it's telling the world by name: I don't trust you, because you have a record.
Third, and the most ruthless: existing models can be used until November 12, but future models — not a single one. The announcement specifically named the upcoming new model Astra, a frontier model tied to "critical network capabilities." The meaning: old accounts can be settled slowly, but don't even think about touching new stuff.
One hand offers full courtesy, the other slams the door shut.
A long record: this is not the first time
OpenAI's "lack of trust" is not baseless. The announcement dug up two old accounts, each with a source.
The first: Twitter. After Musk took over Twitter in 2023, the company violated OpenAI's contract terms. The New York Times reported it at the time. And OpenAI specifically added: we weren't the only ones violated, "it also violated contracts with many other companies." Now Twitter has been folded under SpaceX, and the record follows the entity.
The second, more fatal: the xAI distillation incident. At the end of April this year, Musk admitted under oath during testimony that xAI distilled OpenAI's data to train its own models, directly violating OpenAI's terms of service. The most ironic part: OpenAI pointed out in the announcement that these terms "are similar to xAI's own terms."
The rules you use to demand compliance from others, you yourself are using. And then you admit you violated them.
An entity that admitted in court to violating your terms of service is about to get your model API and can access your unreleased frontier models. If you were OpenAI, would you sign?
Clause dissection: how "Change of Control" became a fatal switch
The thing in this affair most worth studying for all AI practitioners is one clause in that contract: the Change of Control clause.
Routine operation in commercial contracts: when a company is acquired and control changes hands, the partner has the right to reassess or even cancel the agreement within a limited time window. The original intent is to protect the partner from being "sold down the river": you signed with Company A, it was sold to Competitor B, of course I need to reconsider.
OpenAI's custom agreement with Cursor had such a window. And the precision of this operation was very high:
- Exercised the right to cancel, but pushed the cancellation date to the latest allowed by the contract — the interests of existing developers, protected to the last second;
- Simultaneously announced no future models would be provided — new increments, not a single milliliter flows out.
Existing clauses gave OpenAI a completely legal space to act, and the way it chose to act maximized the posture of "legally cutting off supply." You can barely see emotion in the announcement, but every arrangement was calculated.
This is not an ordinary business decision. Business decisions calculate input and output; this one calculated strike precision.
Chain reaction: everyone's next step
What will Cursor do? Just over an hour after the announcement, CEO Michael Truell stepped in. Currently this response has 1.76 million views, 13,000 likes — the likes are double those of OpenAI's announcement itself. He said three things: very regretful; OpenAI models account for only about 5% of Cursor's user traffic, and they are communicating with the OpenAI team to resolve it; the last sentence had a needle wrapped in cotton: "Cursor was one of OpenAI's first customers, we have worked closely with their team for years and trusted their platform as neutral infrastructure for our business."
That 5% figure carries more information than the announcement itself. If true, the actual damage of the cutoff is far less scary than the headline. The main traffic was never on OpenAI models to begin with; users already voted with their feet long ago. But read the other way, the CEO rushing to report such a small percentage in the first hour of a crisis is also standard crisis PR: first make the wound sound small, then discuss next steps.
As for the phrase "neutral infrastructure," the logic of OpenAI's announcement is "I don't trust you," Truell's subtext is "I originally trusted you." One in, one out — this phrase can be considered scrapped today.
Fortunately, Cursor's family assets are still there: the product is still there, the users are still there, the developer mindshare built over four years is still there. Besides, SpaceX owns xAI and Grok — family doesn't talk like strangers — and will likely accelerate the shift to its own models. Whether Grok's coding capability can catch the baton will be seen in the coming months.
What about developers? Counting from today, a 75-day countdown. Those heavily reliant on OpenAI models inside Cursor need to seriously evaluate migration plans: bring your own API Key, switch models, or simply change editors. OpenAI promises to assist with the transition; how much sincerity there is will be seen in the follow-up execution.
As for the industry, this is the real big deal. The AI coding tool space, the hottest startup track of the past three years, has just been dragged directly into the center of a giant war. The last one to receive this kind of treatment was Huawei in the chip industry.
Comment section: a large-scale performance art show
Seven hours after the announcement, the comment section and retweet area had turned into a joke competition.
The most ruthless one didn't curse Musk, but specifically poked OpenAI's sore spot: a company that started as a nonprofit, turned for-profit to chase profit, now runs to its former benefactor's platform to complain about the benefactor, and paid for a blue checkmark certification. "You guys are really too funny."
The meme army entered simultaneously. Someone dug up the classic Spider-Man taking off glasses scene:
Without glasses, OpenAI; with glasses, ClosedAI.
There was an even more cutting three-panel comic: OpenAI solemnly announces "We regretfully decided...", Grok and Cursor opposite give a thumbs up, then go about their business:
But what really carried information in the comment section wasn't the jokes, it was the follow-up from OpenAI's side.
Beyond the announcement, OpenAI technical lead Tibo (@thsottiaux) also posted a response, confirming the partnership termination, citing "trust issues" as the reason, effective date November 12, perfectly matching their own announcement. A netizen heckled below: "Sounds like it's time for another quota reset." This got 1,103 likes, and Tibo himself replied below:
"Resets have their time and place. Soon, but not today."
Translation: compensation is on the way, don't rush.
Among the upvoted comments there was an even more direct one: "It's all good bro, I forgive you. Give us one more reset and we'll call it even." 197 likes:
There was also a comparison image that spread quite widely: the one on the left gives you free quota resets every week, the one on the right takes GPT models from your hands, captioned "Tibo's dark age has begun":
Of course, not everyone was just watching the show. Some had already started seriously calculating.
"One of the reasons I subscribe to Cursor is Codex. Now, I definitely have to reconsider this subscription." No emotion, no jokes, just a thinking emoji. This kind of reaction is the most plain, and also the most deadly — when users vote with their feet, they never warn you in advance.
There was also a comment from a Japanese developer, a few sentences that laid out the stakes very clearly: this news is "quite shocking," bringing your own API Key can indeed still work, but under usage-based billing, not many individual users will bother to go through the trouble. Then he casually dug up an old case: when Windsurf was almost acquired by OpenAI last year, Anthropic did exactly the same thing, directly cutting off Claude's direct supply.
In other words, the model cutoff tactic was not invented by OpenAI. It's just that this time, it was executed most publicly, with the reason stated most bluntly.
Behind the spectacle is actually the same question: who will pay for the developers' losses today?
OpenAI's announcement says "do everything possible to support the transition," technical lead Tibo's hint is "reset is coming soon, but not today"; on Cursor's side, CEO Truell's response is "only 5%, communicating." Both sides are making statements, neither side has yet put real money on the table. In the next two and a half months, the one that submits their answer first takes the goodwill.
Deep revelation: your moat might be someone else's gas station
Stepping outside this feud, this incident is a case that everyone building AI products must remember.
Cursor's story was originally a textbook startup model: the interface is its own, the interaction is its own, the trust of millions of developers is its own. But one thing was always rented — the brain.
The model is rented, which means:
Your growth curve is held in someone else's hands. No matter how well the contract is signed, the Change of Control clause is like a timed switch buried in the foundation: you raise funding, you get acquired, you pick the wrong side — any capital move can trigger it. Cursor wasn't cut off because its product was bad; it was cut off precisely because it succeeded enough to catch Musk's eye.
"Trust" is becoming a prerequisite for model supply. That sentence in OpenAI's announcement, "based on experience, we cannot be confident," marks a shift in model vendors' contract thinking: from "you pay, I supply" to "I need to first confirm whose side you're on." After the xAI distillation incident, frontier model vendors' vigilance toward "will my model become fodder for your training data" will only increase. Access approval for high-sensitivity models like Astra will increasingly resemble arms sale vetting.
Single model dependency = single point of failure, has gone from risk warning to established fact. Previously, advising AI application companies to do model multi-routing was architectural purism; from today onward, it's survival common sense.
At the end of the day, this announcement has drawn a line: all prosperity built atop others' models must answer one question: if tomorrow the model is no longer available to you, what do you have left?
Cursor's next 75 days are the entire industry's public stress test.
Original announcement: Our decision on Cursor following its acquisition by SpaceX (OpenAI official site, 2026-08-29)