OpenAI will cut Cursor off from its models on November 12

OpenAI will stop supplying its models to the code editor Cursor on November 12, 2026. The trigger was ownership: SpaceX completed its purchase of Cursor's parent company, Anysphere, on August 14, 2026 for about $60 billion. OpenAI invoked a change-of-control clause in the supply contract.

Cursor is an AI code editor. Developers use it to have a model read a repository, suggest edits and write code alongside them. It has never trained its own frontier models, so it buys access from several vendors.

Why OpenAI says it is doing this

OpenAI's reason is trust, stated bluntly. It "cannot be confident that SpaceX will use our technology within our terms of service," the company said in a statement reported by IBTimes Singapore. It based that on its experience with Musk's companies violating contracts.

The reporting points at two specific grievances. OpenAI cited past conduct by X, formerly Twitter, and by xAI, Musk's AI company. According to gagadget, the xAI complaint is about distillation, where one company uses another model's outputs to train its own model. That practice generally breaks the supplying vendor's terms.

Musk answered dismissively. bool.dev reports he said he "couldn't care less" and questioned the trustworthiness of OpenAI's leadership.

How much of Cursor actually breaks

Less than the headline suggests. Cursor co-founder Michael Truell puts OpenAI models at about 5% of the product's traffic, a figure all four reports carry.

Truell's framing is the more interesting part. IBTimes Singapore quotes him describing how Cursor had treated OpenAI as "neutral infrastructure," a supplier that sells to everyone regardless of who owns the buyer. That assumption is what actually broke on November 12's schedule. He has said Cursor is in talks with OpenAI to find a resolution.

Reports differ on exactly which OpenAI models are affected, so treat any specific model list with caution. What they agree on is the direction: after the cutoff, OpenAI models are not directly available inside Cursor, while models from other vendors are.

The competitors moved immediately

Anthropic committed to expanding Claude's compute capacity inside Cursor after the cutoff, according to gagadget. Claude, Google's Gemini and Musk's own Grok all remain available in the editor.

That is the competitive shape of this story. A supplier withdrew over an ownership dispute, and rivals moved into the gap within a day. For Cursor, losing 5% of traffic to a vendor while three others expand is survivable. For OpenAI, the calculation is whether the trust signal is worth handing that share to Anthropic.

What this means for developers

Check what your editor is actually calling before November 12. If your team pinned a specific OpenAI model in Cursor for a workflow that matters, that pin stops working. The features most exposed are the ones with no vendor choice in the settings, so audit those first rather than the chat window.

The wider lesson is about supplier risk in your toolchain. Most developers picked Cursor on its editing experience, not on its model contracts. Those contracts turned out to be the load-bearing part, and a corporate acquisition your team had no view of changed them. Any tool that resells someone else's model carries this risk.

Prefer tools that let you supply your own key. Where an editor can route to a model account you control, an upstream contract dispute becomes a configuration change rather than an outage. That is worth more than a small quality difference between models.

Treat "neutral infrastructure" as an assumption to test, not a property to rely on. Truell's word choice names the thing precisely. Model access has been sold like a utility, and this is the clearest case yet of a vendor treating it as a relationship it can end. Plan for at least two model providers in anything you would be unhappy to see stop working in ten weeks.