Reading AI status pages
AI providers split their status by surface. OpenAI and Anthropic each report the consumer app and the developer API as separate components, and an incident commonly touches one model or one feature rather than the whole product. That is why the overall state can read "minor incident" on a day when your own chats work normally.
Tools built on top of these APIs inherit their incidents. Cursor, for example, sends requests to several model providers; when one of them has elevated errors, the editor slows down for that model only. Seeing the providers next to each other on this page makes that pattern easy to spot.
What a declared incident usually means
Most incidents on these pages are described as elevated error rates or degraded performance and are resolved within the hour. The declared start and end of every incident are kept on each service page, together with the text of each official update, so you can check how long a past problem really lasted instead of relying on memory.