OpenAI announced on 28 August that it will cut Cursor off from its models entirely from 12 November 2026 — a direct consequence of SpaceX completing its $60bn acquisition of the AI coding tool in mid-August. OpenAI’s stated reasoning, via spokesperson Thibault Sottiaux, “boils down to trust”: the company cited Elon Musk’s history of breaking contracts elsewhere, and framed the decision partly as a safety call given OpenAI’s forthcoming “Astra” model carries advanced cyber capabilities that demand a higher bar of accountability from partners with access to it.
Cursor CEO Michael Truell downplayed the practical impact — OpenAI models account for only around 5% of Cursor’s user traffic today, since the product already routes to Anthropic, Google and xAI models — and said he’s confident of “finding a solution.” What’s more telling is who moved fastest to fill the gap: Anthropic co-founder Tom Brown publicly reaffirmed Cursor as “a trusted partner of Anthropic since Sonnet 3.5” and committed to expanding Claude compute capacity for Cursor specifically. Musk’s own response — “I couldn’t care less” — suggests OpenAI’s model access was never central to Cursor’s product strategy in the first place.
Why this matters if you’re not the one writing code
Most coverage of this story treats it as tech-industry gossip about a billionaire feud. For anyone commissioning software, it’s something more useful: a live, public case study in what happens when a coding tool’s model access sits outside its own control. Cursor’s engineering team didn’t choose to lose OpenAI access — a corporate ownership change three levels removed from the product decided it for them, on a deadline. Multi-model routing (which Cursor already had, and which is exactly why the practical damage here is limited) turns out to be the thing that made this survivable rather than existential.
That’s the pattern worth taking away, independent of which vendor wins this particular round: a development team’s AI coding stack carries a dependency risk that sits above the tool itself, in who owns it and who controls the model relationships behind it. A single-model, single-vendor setup is a bet on that vendor’s corporate stability and contractual goodwill holding for the life of the project, not just its current feature set.
So what
When we scope AI-assisted development work, part of that scoping is which tools and model providers a build depends on, and how exposed that dependency leaves the client if a vendor relationship changes overnight — not treating tool choice as a developer preference to be settled after the contract is signed. See how we approach this in AI-assisted development, or get in touch if you want that risk factored into your project from the start.