OpenAI, Anthropic and xAI went down within hours of each other and none of them named a shared cause

WIRED reported on 3 September that frontier models from Anthropic, OpenAI and xAI all had outages that same morning, taking down their chatbots for a period. The piece, by senior writer Lily Hay Newman, is built around the fact that nobody involved named a shared cause.

The timings, as WIRED reports them, are close enough to look connected. Anthropic began warning about a partial outage at 6:23 am PT, covering elevated errors on requests to Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5. It said shortly after that it had identified the cause and deployed a fix, and marked the issue resolved at 9:16 am PT. Claude Sonnet 5 appeared to have similar trouble briefly after 9 am PT. xAI posted "investigating outage" on its status page at 6:30 am PT and marked the episode complete at 10:05 am PT.

OpenAI gave WIRED a specific account through spokesperson Kathleen Chaykowski: a routing error starting around 7:43 am PT made ChatGPT and Codex unavailable for some users across platforms, with a fix implemented by about 8:17 am PT.

The shared-cause theory has nothing behind it

WIRED's own framing is that simultaneous failures in one sector would normally point at a cloud provider, a content delivery network or another shared vendor. That explanation did not survive contact with the evidence. WIRED reports that Cloudflare, Amazon Web Services and Microsoft Azure recorded no outages that day, and that neither OpenAI nor Anthropic cited an external source.

The one company that did name something was SpaceX, xAI's parent, which said the Grok issues came from an outage at its Memphis compute centre that morning and added an apology to its compute partners. WIRED notes that Anthropic and xAI announced a compute partnership with SpaceX in May. That is a thread rather than an answer: Anthropic declined to comment on the episode, and SpaceX did not respond to WIRED's request for comment.

Correlated failure is the planning assumption now

For a team running anything on these APIs, the useful part is the shape of the window. Between roughly 6:23 am and 10:05 am PT, three of the largest model providers were degraded at once, and two of them still have not explained why in public.

Most retry logic is written for the opposite case. It assumes the provider is having a bad minute and that waiting will fix it, so it backs off and tries the same endpoint again. In a window like this that loop just burns the request budget and returns the same error. A fallback that actually helps has to reach a different provider, which means the prompt format, the tool definitions and the output parsing have to already work against a second model before the outage starts. Deciding that during the incident is too late.

The second thing worth taking concerns what a status page actually tells you. Two of these three providers marked their incidents resolved without ever saying what happened, so a team writing its own postmortem for that morning has no upstream account to reference.

Source

Nobody Is Saying Why OpenAI and Anthropic Had Outages Today, WIRED, 3 September 2026.