Why Did OpenAI Ban Cursor's Models After the SpaceX Acquisition?
OpenAI is pulling Cursor's native model access after SpaceX's acquisition, citing distillation risk and Elon Musk's history of contract violations.
What happened between OpenAI and Cursor?
OpenAI told SpaceX it plans to wind down the contract that lets Cursor offer OpenAI’s models natively inside its coding platform. The announcement went out in a blog post published on a Friday evening, a timing choice companies typically use when they want news to get as little attention as possible. OpenAI’s stated reason wasn’t pricing or competition in the abstract. It was trust: the company said it “cannot be confident that SpaceX will use our technology within our terms of service,” pointing directly to Elon Musk’s track record of contract violations at his other companies.
This matters because Cursor, until recently an independent company, was acquired by SpaceX (through xAI) a few months before this announcement. That acquisition is what triggered OpenAI’s decision, and it’s worth understanding both the business logic and the long-running feud behind it.
TL;DR
- OpenAI is cutting off Cursor’s native access to its models, meaning developers will no longer get GPT models built directly into the Cursor interface without supplying their own API key.
- The stated reason is distillation risk: OpenAI worries that data flowing through Cursor could be used to train competing models cheaply, violating its terms of service.
- OpenAI explicitly cited Elon Musk’s history of broken contracts, including sworn admissions that xAI violated OpenAI’s terms of service around distillation.
- The dispute traces back to the 2018 OpenAI/Musk split and the ongoing lawsuits and public feud between Musk and Sam Altman since xAI launched in 2023.
- Cursor’s CEO Michael Truel pushed back, saying OpenAI models account for only about 5% of Cursor’s user traffic, a figure that likely reflects token volume rather than revenue.
- This is not an isolated incident: Anthropic previously cut off OpenAI and xAI from Claude access, and restricted Windsurf after Google acquired it, for similar distillation concerns.
- Developers can still use OpenAI models in Cursor by manually adding their own API key, but the built-in, no-setup access is going away.
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Why is OpenAI worried about Cursor specifically?
The core concern is distillation. Distillation is a well-established machine learning technique where a smaller “student” model is trained on question-and-answer pairs generated by a larger “teacher” model. Instead of collecting and cleaning training data from scratch, a company can just query a frontier model repeatedly and use its outputs to train a cheaper model that approximates its behavior. Done at scale, this is sometimes called a distillation attack, and Anthropic has publicly accused Chinese AI labs of running industrial-scale versions of it against Claude, spinning up huge numbers of accounts to harvest responses.
OpenAI’s worry is that once SpaceX owns Cursor, every prompt and response flowing through Cursor’s OpenAI integration becomes a potential training resource for a SpaceX-controlled AI effort, whether that’s xAI’s Grok models or something built specifically around Cursor’s coding data. Cursor sits on a large volume of real-world coding interactions. Pairing that data with xAI’s compute buildout (Colossus and Colossus 2, its large Nvidia GPU clusters) gives a competitor a plausible shortcut to a strong coding model without the years and dollars OpenAI spent getting there.
What does Elon Musk’s history have to do with it?
OpenAI’s blog post didn’t just gesture vaguely at risk. It named specific incidents. It said Musk’s companies broke the terms of a prior contract after acquiring Twitter, and that under oath earlier this year, Musk admitted xAI had violated OpenAI’s terms of service specifically around distillation. That admission is the crux of OpenAI’s justification: this isn’t a hypothetical worry, it’s citing a documented instance of the exact behavior it says it’s trying to prevent.
The backstory runs deep. OpenAI was founded in 2015 as a nonprofit, with Musk among its early funders. Around 2017 and 2018, OpenAI decided it needed serious capital to compete with Google DeepMind, which led to a structural shift and a falling out: Musk reportedly wanted to run the company, Sam Altman said no, and Musk left. OpenAI later adopted a capped-profit structure in 2019, then ChatGPT’s 2022-2023 breakout turned it into one of the most valuable private companies in the world. Musk launched xAI as a direct competitor in 2023, then sued OpenAI in 2024 over its shift from nonprofit to a structure capable of generating enormous returns for investors. That suit was dismissed, but the public back-and-forth, including released private emails, left plenty of residual animosity. Against that backdrop, OpenAI handing its models to a company now owned by Musk, with no guardrails, was apparently a risk it wasn’t willing to take.
How did Cursor respond?
Cursor’s CEO, Michael Truel, posted that he was “sorry to see” OpenAI’s move and said the company was in talks to resolve it, though a walk-back seems unlikely given how pointed OpenAI’s language was. More notably, he offered a number meant to downplay the impact: OpenAI models, he said, serve only about 5% of Cursor’s user traffic.
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
That figure drew a quick rebuttal from Thibault “Tibo” (a well-known voice in the coding-AI space), who argued tokens aren’t a good proxy for revenue or value. OpenAI’s models sit near the frontier of token efficiency, meaning they need fewer tokens to accomplish the same task compared to smaller or weaker models. A low share of total token volume doesn’t necessarily mean a low share of spend or a low share of the work that actually matters. Weaker or open-source models can rack up token counts on simpler tasks while frontier models handle the harder, more valuable ones with less overhead. In other words, 5% of tokens might represent a much larger share of Cursor’s actual value delivered, or of OpenAI’s revenue from the relationship.
Is this actually unprecedented?
No, and that’s part of what makes this story more than a one-off spat. Similar moves have happened repeatedly over the past year or so:
Google acquired Windsurf, another AI coding platform that, like Cursor, didn’t build its own frontier models and instead relied on providers like Anthropic and OpenAI. Shortly after, Anthropic began limiting Windsurf’s direct access to Claude models, a move reported by TechCrunch. The logic was the same one OpenAI is now applying to Cursor: once a rival owns the platform, handing it unrestricted model access looks less like customer support and more like feeding a competitor.
Anthropic had already revoked OpenAI’s direct access to Claude models over similar distillation concerns, and separately cut off xAI from Claude models months before the Cursor news broke. So by the time OpenAI made its announcement about Cursor, the pattern of frontier labs restricting rivals’ access to their models, especially after an ownership change tied to a competitor, was already established. OpenAI’s move against Cursor fits a broader industry trend of AI labs treating model access as a strategic asset to be guarded, not just a product to be sold.
What happens next for developers using Cursor?
Developers won’t lose OpenAI models in Cursor entirely. The built-in, pay-through-Cursor access is going away, but users can still bring their own OpenAI API key and configure it manually. That’s a meaningfully different experience: it shifts billing and account management to the developer directly with OpenAI, rather than bundling it invisibly into a Cursor subscription. For teams already juggling multiple model providers across coding tools, this is an annoyance rather than a dealbreaker, but it does remove some of the convenience that made Cursor attractive as a one-stop shop across Claude, Gemini, and GPT models.
Frequently Asked Questions
Why did OpenAI ban Cursor from using its models?
OpenAI cited concerns that SpaceX, which acquired Cursor, could use access to OpenAI’s models to train competing AI systems through distillation, a practice OpenAI said Musk’s companies have already violated contracts over.
Can Cursor users still access GPT models at all?
Yes, but not natively. Users can add their own OpenAI API key to Cursor and continue using GPT models that way, rather than through Cursor’s built-in, bundled access.
What is a distillation attack in AI?
It’s the practice of repeatedly querying a large “teacher” model to collect its outputs, then using those question-answer pairs to train a smaller “student” model cheaply, effectively copying much of the teacher model’s capability without the original training cost.
Has this kind of model access restriction happened before?
Yes. Anthropic previously limited Windsurf’s access to Claude after Google acquired it, cut off OpenAI’s direct access to Claude, and restricted xAI’s access to Claude models months before OpenAI’s move against Cursor.
How much of Cursor’s usage actually depends on OpenAI models?
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
Cursor’s CEO said OpenAI models account for about 5% of user traffic, though critics point out that token share doesn’t necessarily reflect revenue or actual value delivered, since frontier models tend to need fewer tokens per task than smaller models.