Claude went down for an hour on Monday, and messages sent between 14:00 and 14:59 UTC may not have been saved

If you were using Claude on Monday afternoon and a chat looked like it sent, it may not have. Anthropic recorded a one-hour incident on the Claude status page that hit claude.ai, Claude Code, Claude Cowork, the Claude Console and the Claude API, and its own note says some messages sent between 14:00 and 14:59 UTC may not have been saved.

The first entry on the incident is at 14:21 UTC. Anthropic said Claude.ai, its desktop and mobile apps, Claude Code and Claude Cowork were returning elevated error rates, that users might see failed requests, errors loading or sending conversations, or be asked to sign in again, and that a retry might succeed. Some Claude API requests were failing too. The affected components listed at the bottom of the page are claude.ai, Claude Console at platform.claude.com, the Claude API at api.anthropic.com, Claude Code and Claude Cowork.

What broke, and when

At 14:41 UTC Anthropic said it had applied a mitigation and error rates had dropped a lot, and that sign-in attempts, account actions and Claude Code and Cowork sessions were still failing. At 15:00 UTC Anthropic added that the errors that began at 14:00 UTC had been mitigated at 14:36 UTC and that existing conversations were mostly working again, but that a second issue was preventing many people from signing in. Single sign-on and Sign in with Apple were unavailable, and starting new chats, voice conversations, Claude Code and Cowork sessions, purchases and file uploads were all failing. The update ended with a specific instruction: if you were signed in, do not sign out.

At 15:11 UTC Anthropic said most services had recovered as of 14:59 UTC and repeated the message-loss note: "Some messages sent between 14:00 and 14:59 UTC may not have been saved." At 15:30 UTC it said all services had been operating normally since 14:59 UTC and that it was continuing to monitor. There is no postmortem on the page yet.

What is worth checking

For anyone running a Claude Code session during that window, the chat history is not a safe copy of what the session did. A file edit or a shell command that ran, and then failed to write to the transcript, will not show up when you scroll back. If a session ended by asking you to sign in again, the messages inside it are the most likely ones to be missing. Comparing the working tree, the git log, or any test output against what the transcript claims is a five-minute check, and it is the difference between building on real work and building on work the transcript does not actually record.

API calls in the same window have to be judged the same way. A 5xx that arrived is a signal to retry. A 200 that arrived with content is safe. A request that did not return, or that returned late, is worth reconciling against whatever the caller stored on its own before assuming the call went through.

Source

Primary: Anthropic status page: Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API.