Claude went down this morning, taking Claude Code and the API with it

If your Claude Code session failed this morning, it was not your setup. Anthropic suffered a widespread outage on Tuesday that knocked out claude.ai, Claude Code, Claude Cowork, the Claude Console, and parts of the Claude API for most of an hour, and the company warns that some messages sent during the window may never have been saved.

The trouble began around 14:00 UTC, which is 7 a.m. Pacific. Anthropic's status page opened an incident at 14:21 UTC, reporting elevated error rates across claude.ai, including the desktop and mobile apps, plus Claude Code and Claude Cowork. Users saw failed requests, conversations that would not load or send, and prompts to sign in again. By 14:28 UTC, some Claude API requests were failing too, and by 14:30 UTC the company had added the Claude Console to the affected list and was describing the situation as degraded performance across the board.

A mitigation applied at 14:41 UTC brought error rates down substantially, but then came a second wave. Single sign-on and Sign in with Apple went unavailable, along with new chats, voice conversations, purchases, and file uploads. The status page's advice during the second wave: if you are signed in, do not sign out. Downdetector logged a peak of more than 18,000 outage reports, against a normal baseline near zero. By 15:30 UTC, Anthropic posted that all services had been operating normally since 14:59 UTC, and moved the incident to monitoring. No root cause was disclosed.

The practical fallout matters more than the timeline. Anthropic warned that some messages sent between 14:00 and 14:59 UTC may not have been saved. If you were mid-session in Claude Code this morning, it is worth checking that your conversation history is intact and that any work you described in chat made it into the transcript. The "account temporarily unable to authenticate" error many users saw suggests the failure touched the identity layer, not just inference capacity, which would explain why sign-ins broke while some retries succeeded.

A bad week for reliability, and a crowded news day

The outage comes one day after Claude Sonnet 5.5 became the default Sonnet model on Anthropic's API. Anthropic has not connected the two events, and there is no evidence the model rollout caused the outage, but the timing will invite speculation until the company says more.

It also lands on a day when the whole industry is talking about agent reliability. OpenAI spent DevDay selling agents that run overnight in the cloud, while separately cancelling a model for crossing authorization boundaries. Anthropic, meanwhile, had a morning where its own flagship tools could not be reached at all. Reliability is the boring foundation underneath every ambitious agent pitch, and Tuesday was a reminder that the foundation still cracks.

For developers, the takeaway is the familiar one: build your workflows so an hour-long provider outage is an annoyance, not a disaster. Keep local copies of anything important, and treat cloud-hosted agent sessions as something that can vanish mid-task.

Sources

Unite.AI on the status page timeline · AI Weekly outage alert · PC Guide · SQ Magazine