The Tap on the "Borrowed Brain" Runs Dry — Two Weeks After SpaceX Bought Cursor, OpenAI Serves Notice to Cut Off GPT Access
Two weeks after SpaceX acquired Cursor, OpenAI served notice that it would cut off access to its GPT models, with a planned shutoff on November 12. We break down — from a developer's perspective — the features a personal API key can't cover and Anthropic's swift move to expand its own capacity.
A Promised Model, Pulled in Two Weeks
SpaceX acquired Anysphere, the company behind the coding editor Cursor, for roughly $60 billion on August 14, 2026. Just two weeks later, on August 28, OpenAI invoked the contract's change-of-control clause and served notice that it would cut off GPT model access to Cursor. The planned shutoff date is November 12, though OpenAI added that this is "not final and could come sooner."
Why OpenAI Is Cutting Off Supply
In an official post, OpenAI explained that it could not be confident SpaceX would use its technology within the bounds of OpenAI's terms of service. Citing past instances in which Musk-affiliated companies allegedly violated those terms, OpenAI chose to terminate preemptively rather than wait for an actual breach. It has also been reported that the company's prioritization of internal inference compute toward safety evaluations factored into the decision.
In other words, this cutoff was triggered not by performance or technical merit but by "who owns the company." Model supply itself has started to be used as a bargaining chip to keep a competitor in check.
What Stops, and What Doesn't
Some see the impact as limited. According to Cursor CEO Michael Truell, OpenAI models account for only about 5% of Cursor's usage traffic, with the rest going to Claude, Gemini, and Grok. But "just plug in your own key and you're fine" doesn't always hold. A personal OpenAI API key only powers local Chat and Agent; the right-hand column in the table below can't be covered that way.
| Can be replaced with a personal API key | Cannot be replaced (depends on Cursor's supply) |
|---|---|
| Local Chat | Tab (code completion) |
| Local Agent | Auto |
| — | Cloud / Background Agents |
| — | Automations |
| — | CLI / API / SDK |
Everything in the right-hand column is a feature where Cursor's own infrastructure supplies and routes the models — in effect, the "self-driving" part of the editor. The more a team built these around GPT models, the greater the effort to switch.
Anthropic Moved Instantly
On the other side of the shrinking supply, Anthropic immediately moved to expand its own capacity. On August 29, co-founder Tom Brown announced that the company would "add compute to power Claude models in Cursor." It's a play to grab the freed-up 5%, and here again you can see model supply becoming a weapon of competition.
The Risk of Resting Your Footing on a "Borrowed Brain"
The lesson for developers is that you now have to look not just at a tool's performance but also at the stability of its supply. The editor may sit on your machine, but its brain is borrowed from another company — and the scenario where that lender tightens the tap over an acquisition or a reading of the terms of service has become real. Beneath the convenience, a setup that depends heavily on a single company's model carries this kind of risk.
Guarding against it isn't hard for now. Check which model you're currently using in Cursor's settings, try your main work with Claude or Gemini, and decide on a migration target before November 12. If you've built automations that assume a specific model, it's safer to have an alternate route ready. This episode shows that "where you source your model from" has become part of workflow design.
References: Forkast / XenoSpectrum / Bitroot / The New Stack / Digital Trends
