What's this about?
The blog post lays out the case in order. OpenAI relies on custom contracts with large partners to ensure terms-of-service compliance and safety at scale. It cites two prior incidents: after Musk acquired Twitter — now part of SpaceX — the company broke the terms of OpenAI's contract alongside many others; and under oath in April 2026, Musk admitted xAI, now also part of SpaceX, had distilled OpenAI outputs to train its own models, which OpenAI's terms prohibit. The custom Cursor agreement contained a change-of-control cancellation window. OpenAI used it, chose the latest permitted date, and stopped supplying future models immediately.
What is model distillation, and why is it prohibited?
Training your model on another model's outputs. It works remarkably well — the student inherits a large share of the teacher's capability at a fraction of the training cost — which is why every major provider's terms forbid it. It is the cheapest way to convert someone else's several-hundred-million-dollar training run into your product, and it is nearly impossible to detect from the outside, which is why an admission under oath matters so much more than a suspicion.
What's the community saying?
- Chinese-language commentators noted the symmetry: within two hours of the cutoff, Anthropic publicly committed more compute to Cursor.
- The post also names Astra as a reason — OpenAI says it now has "a new level of accountability" over how upcoming models are used.
- The most common practical question was whether bring-your-own-key survives, which Cursor's customer email does not clearly answer.