OpenAI Cuts Cursor Off From Its Models on Nov 12
OpenAI invoked a change-of-control clause after SpaceX closed its acquisition of Cursor, ending model access on November 12. Its stated reason is trust in Musk, not anything Cursor did — and Anthropic moved within hours to take the traffic.
OpenAI told SpaceX on August 28 that it is winding down the contract that supplies its models to Cursor, with the shutoff set for November 12 — the longest notice period the agreement allows. The trigger was ownership: SpaceX’s all-stock acquisition of Anysphere, Cursor’s parent, closed on August 14, and OpenAI’s contract carried a change-of-control clause with a limited cancellation window attached to it.
OpenAI’s stated reason is not performance or price. It is trust. “It boils down to trust,” said OpenAI’s Thibault Sottiaux, who framed the decision around the company’s history with Elon Musk’s other businesses rather than anything Cursor has done. OpenAI pointed to Musk cancelling its $2 million-a-year licence for Twitter data after taking over that company, and to Musk’s own admission that xAI trained Grok on output from other labs’ models — a direct terms-of-service breach. It also invoked the cyber capabilities of its forthcoming Astra model as a reason to hold partners to a higher standard of accountability.
The practical blast radius is smaller than the headline. Cursor co-founder Michael Truell said OpenAI models account for roughly 5% of Cursor’s AI traffic — most users are on Cursor’s in-house Composer models or on Anthropic’s. Two escape hatches also survive the cutoff: developers can keep using GPT models inside Cursor with their own OpenAI API keys, and OpenAI’s own IDE extension continues to work inside Cursor. What ends is the commercial arrangement in which Cursor serves OpenAI models to its users as a first-party option, and OpenAI has said Cursor will not receive future models during the wind-down.
Truell said he was surprised, noting that Cursor was “one of the very first users” of OpenAI’s technology, and that his team is talking to OpenAI to find a resolution. Musk was less interested in one: asked about the cutoff, he said he “couldn’t care less.”
Anthropic moved within hours to occupy the gap. Tom Brown publicly committed to keeping Cursor supplied — a notable position from the lab that cut Windsurf’s API access during its own acquisition drama, and that has restricted OpenAI’s access before. Whichever way the principle is argued, the commercial logic is not subtle: Cursor is one of the largest single buyers of frontier coding tokens in the market, and OpenAI has just vacated 5% of it.
The wider signal is about what a model licence now is. Frontier labs have spent two years selling API access as neutral infrastructure. Three cutoffs in eighteen months — Anthropic to Windsurf, Anthropic to OpenAI, now OpenAI to Cursor — establish something closer to the opposite: model access is a commercial relationship that can be revoked when the counterparty’s ownership changes, and change-of-control clauses are the mechanism. Any company building on a single lab’s models now has an acquisition risk it did not price in, and it is not the acquirer’s risk. It is the target’s customers’.
More on Cursor
Evergreen coverage we keep current — start here.
Want AI news before everyone else?
The morning's most important AI stories, straight to your inbox. No fluff.