OpenAI's GPT-5 models exit Cursor Nov 12 — line up a fallback
The SpaceX-triggered cutoff is 9 days out, and a personal API key won't cover background agents. Plus: Anthropic's S-1 flags a self-improvement threshold.
Copy markdown
SpaceX bought Cursor, so OpenAI is cutting it off
OpenAI confirmed it's ending Cursor's direct access to its models on November 12 after SpaceX acquired Cursor's parent, invoking a change-of-control clause. Only OpenAI's own models (the GPT-5 line) go dark — Anthropic, Google, Moonshot, Grok and Cursor's own Composer keep working, and OpenAI puts those models at roughly 5% of Cursor traffic. If any agent, task or rule in your workflow is pinned to a GPT-5 model, repoint it before the 12th.
Your own OpenAI key only plugs half the hole
After the cutoff you can bring a personal OpenAI API key (or an Azure/Bedrock gateway) into Cursor, but it only works in local Chat and Agent — not Tab, Auto, Cloud or Background agents, Automations, the CLI, or Cursor's API/SDK, which keep using the models Cursor supplies. Usage bills to your API account at list prices (a ChatGPT subscription doesn't cover it), so price out your heavy agent runs now and decide what moves to Claude, Gemini or Composer.
Anthropic tells the SEC it's near self-improving AI
In an amended S-1 filing, Anthropic reportedly warned it may imminently cross its internal recursive-self-improvement threshold and conceded it has sometimes recognized big capability jumps only after the fact — floating possible Responsible Scaling Policy updates to keep humans supervising R&D. For teams building on Claude, it's a signal to watch for RSP changes that could reshape how and when new models ship.