Four major AI models suffer rare overlapping downtime
Ars Technica Kyle Orland
OpenAI, Anthropic, xAI and Google all hit outages Thursday morning. Rare for this many big AI systems to wobble at once.
Based on reporting by Ars Technica, Kyle Orland — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A weird one hit the AI world Thursday morning: cloud models from OpenAI, Anthropic, xAI and Google all ran into significant service trouble over the same stretch of hours. The overlap was unusual enough to stand out on status pages, even if the companies were dealing with their own separate problems.
Anthropic was first to flag trouble, calling it a partial outage at 9:23 a.m. Eastern. The company said requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 were producing elevated errors. About 15 minutes later, Anthropic said it had identified the cause, and by 12:16 p.m. it said a fix had been deployed and the issue was resolved.
There was another Anthropic hiccup after noon, this time tied to Claude Sonnet 5. That one was brief, but it added to the sense that the morning was not going especially smoothly for the company’s models.
OpenAI had its own problem starting at 10:43 a.m., when it reported elevated errors across ChatGPT and Codex and said users were seeing degraded performance. A mitigation went in a little more than half an hour later, and by 12:55 p.m. OpenAI marked the issue resolved.
The source doesn’t spell out the exact failures at xAI or Google, only that they were part of the same rare cluster of downtime. Still, the headline is simple: four major AI providers had serious service interruptions on the same morning, which is a neat reminder that “always on” is still more promise than reality.
My take — AI-written commentary, not fact-checked reporting
The AI industry loves talking about scale, but boring uptime is the real product and too many companies treat it like an afterthought. When four giants wobble on the same morning, the mystery isn’t that one model hiccuped; it’s that anyone still calls these systems dependable enough for the center of the stack.
Read more about this at: Ars Technica