Claude is down: what's failing and what you can do

Update (15:30 UTC): all services have been working normally since 14:59 UTC. Anthropic is still monitoring in case it fails again. If you sent messages between 14:00 and 14:59 UTC, check they were saved: some may have been lost.

On September 29 Claude spent about an hour returning errors across almost its whole ecosystem: the web, the apps, the API, Claude Code and Cowork. If it asked you to sign in again or Claude Code hung, it wasn't your machine.

◢What happened on September 29, 2026

• ▹15:30 UTC, everything back to normal: Anthropic confirms all services have been operating normally since 14:59 UTC and keeps monitoring.
• ▹15:11 UTC, service restored: signing in, new chats, voice conversations, Claude Code and Cowork sessions, purchases and file uploads work again. Some messages sent between 14:00 and 14:59 UTC may not have been saved.
• ▹15:00 UTC, the cause becomes clearer: the errors began at 14:00 UTC and were mitigated at 14:36 UTC. A second issue was preventing many users from signing in (single sign-on and Sign in with Apple were down), starting new chats, using voice, starting Claude Code or Cowork, making purchases and uploading files. Anthropic explicitly asked signed-in users not to sign out.
• ▹14:41 UTC, mitigation applied: error rates drop substantially, but some sign-ins, account actions and Claude Code and Cowork sessions are still failing.
• ▹14:30 UTC, scope expands: Anthropic confirms the degradation also affects the Claude API and the developer platform.
• ▹14:21 UTC, investigation starts: Anthropic investigates elevated error rates and notes that retrying requests may succeed intermittently.

◢Affected services

• ▹Claude.ai (web, desktop and mobile)
• ▹Claude Platform and API (platform.claude.com and api.anthropic.com)
• ▹Claude Code
• ▹Claude Cowork

Claude for Government was not affected.

◢How to check if Claude is down

The reliable source is the official status page, status.claude.com. Anthropic posts every incident there with timestamps and the status of each service separately: claude.ai, the console, the API, Claude Code and Cowork. You can subscribe to get an email when there's news.

If the status page is all green and it's still failing for you, then it's worth checking your side: your connection, a VPN, a browser extension or the Claude Code version you have installed.

◢What to do if Claude goes down

• ▹Don't sign out or reinstall anything. If sign-in is what's broken, signing out only locks you out until it's fixed. In this outage it's exactly what Anthropic recommended.
• ▹Retry after a while. In most incidents, retrying works intermittently before the service is fully back.
• ▹Keep important work outside the chat. If you're writing something long during an outage, copy it before sending: messages from that window may not be saved.
• ▹If you work with the API, retry with growing waits. An outage like this is exactly why any serious integration should retry with exponential backoff instead of failing on the first error.
• ▹Have a plan B for production. Claude models are also served through Amazon Bedrock and Google Cloud Vertex AI. Having one of them set up as a fallback can let you keep your service running when direct access fails.

◢Recent outages

• ▹September 22, 2026: elevated errors on Claude Mythos 5.1, Fable 5.1 and Opus 5 between 00:50 and 02:10 UTC. Resolved.
• ▹September 16, 2026: issues with subscriptions through Google Play, caused by a Google outage. Resolved.
• ▹September 15, 2026: intermittent error spikes on Claude Mythos 5.1 and Fable 5.1. Resolved with a fix.

If your company already relies on AI to get work done, an outage like this is a good reminder to use it wisely and have alternatives ready. It's one of the things I cover in my AI training for companies.