OpenAI Plans to Cut Cursor Off, So Developers Create Emergencies

OpenAI plans to end Cursor’s access to its AI models on November 12, after SpaceX acquired the coding startup, while Cursor says OpenAI models account for about 5% of its traffic and is negotiating a resolution. A contingency build assembled during the dispute has now assigned that remaining share to code that is already in trouble.

The build only calls the OpenAI model when an automated test fails. A green check mark receives one of Cursor’s other models; a red failure opens a narrow rescue channel. The arrangement was meant to preserve a small amount of access without pretending the contract dispute had gone away.

The emergencies became deliberate

Developers quickly discovered that a harmless failing test was enough to summon the model. They began adding one to otherwise finished projects, then asking for help with the emergency they had just created. When the model fixed it, the channel closed, forcing the team to introduce a second failure before the next request.

Our build is failing in exactly the way that keeps it alive.

A developer testing Cursor’s contingency build

The practice spread to release pipelines, where teams started marking the deliberate failures as accepted exceptions. A clean build was no longer a sign that the software was ready; it was evidence that nobody on the project could reach the model reserved for urgent work.

By November 11, several projects are expected to have one carefully maintained failure between the final test and deployment. Engineers say removing it would produce the most dangerous result the contingency was designed to avoid: code that works, and no emergency at all.