Claude had a global outage on July 17, 2026, driven by “529 Overloaded” server errors that hit the API, Claude Code, the web app, and the desktop app. It lasted about 1 hour and 32 minutes. StatusGator detected it and sent an Early Warning Signal at 14:30 UTC, 27 minutes before Anthropic acknowledged it at 14:57 UTC.
Incident at a glance
| Detail | Value |
|---|---|
| Service | Claude (Anthropic) |
| Date | July 17, 2026 |
| Status | Resolved |
| Duration | ~1 hour 32 minutes |
| Primary symptom | 529 Overloaded / server overload |
| Affected surfaces | API, Claude Code, claude.ai web app, desktop app |
| First user report | 13:43 UTC |
| StatusGator Early Warning Signal | 14:30 UTC |
| Provider acknowledged | 14:57 UTC |
| Recovery | 15:15 UTC |
| Detection lead | 27 minutes before acknowledgment |
Was Claude down on July 17, 2026?
Yes. Claude experienced a global outage on July 17, 2026, starting around 13:43 UTC and recovering by 15:15 UTC. The first signs were “529 Overloaded” errors and incomplete responses, and within an hour reports had arrived from six continents. The issue did not appear to affect every single user at once, but it was widespread and touched every major Claude surface.
What caused the Claude outage?
The outage was a server-overload event. The dominant error users saw was HTTP 529 “Overloaded,” which Claude returns when its servers cannot keep up with demand. That single condition produced a range of downstream symptoms:
- 529 Overloaded errors on the API and in Claude Code
- Incomplete or cut-off responses
- “Taking longer than usual” retry loops that eventually failed
- claude.ai not loading and the Windows desktop app not working
StatusGator classified the incident as Claude Code requests failing due to server overload, though the reports show the overload reached well beyond Claude Code.
How long was Claude down?
The outage lasted roughly 1 hour and 32 minutes, from the first user report at 13:43 UTC to the last one at 15:15 UTC. Anthropic acknowledged the incident partway through, at 14:57 UTC, and service recovered shortly after.
Timeline of the July 17 Claude outage (UTC)
| Time (UTC) | Event |
|---|---|
| 13:43 | First outage reports reach StatusGator |
| 13:43 to 14:29 | Reports build: 529 Overloaded errors, incomplete responses, and Claude Code failures across the Philippines, Ukraine, Germany, and Argentina |
| 14:30 | StatusGator sends an Early Warning Signal to subscribers |
| 14:30 to 14:56 | Report volume climbs across North America, Europe, the Middle East, and Asia-Pacific |
| 14:57 | Anthropic acknowledges the incident on its status page, 27 minutes after StatusGator’s alert |
| 15:15 | Final outage reports arrive; recovery |
Who was affected?
The overload cut across the entire product, not one feature. Because the failure was at the server level, developers on the API, Claude Code users, and people using the web and desktop apps all hit it at once. Reports arrived from the United States, Canada, the UK, Germany, France, Spain, Italy, Sweden, Ukraine, Jordan, Morocco, India, the Philippines, Hong Kong, Australia, Argentina, and Peru within the first hour.
A snapshot of what users reported, from around the world:
- “529 Overloaded” (Košice, Slovakia)
- “The server is overloaded on claude chat” (Mörfelden-Walldorf, Germany)
- “claude code not responding” (Hong Kong)
- “Not loading the site. Claude.ai” (Kerala, India)
- “Taking longer than usual (attempt 9)” (Santa Rosa Beach, Florida)
- “not working, no output” (Cebu City, Philippines)
- “the server is overloaded please try again later” (Amman, Jordan)
How did StatusGator detect the outage early?
StatusGator flagged the outage at 14:30 UTC, 27 minutes before Anthropic’s status page did. It works by combining official status feeds with real-time user reports, so a spike in unofficial reports can trigger an Early Warning Signal before the provider confirms anything.
That head start matters. Counting from the first user report at 13:43 UTC, the crowd knew 74 minutes before the incident was official. Subscribers who received the 14:30 signal used that window to pause automations, fail over to another model or provider, and warn their teams while Anthropic was still investigating.
Two patterns stood out in the data:
- The reports self-diagnosed the incident. Users named the exact failure mode (529 Overloaded, server overload, retry loops) long before any official root cause.
- It hit everything at once. The API, Claude Code, claude.ai, and the desktop app failed together, worldwide, inside an hour. That is exactly the kind of broad, fast event where minutes of early warning matter most.
How to stay ahead of the next Claude outage
- Do not wait for the official acknowledgment. It lagged the first user reports by more than an hour here. If you watch only the provider’s page, you inherit that delay.
- Treat 529s and retry loops as a leading indicator. Overload errors and “taking longer than usual” retries are early symptoms of a broader outage, not isolated hiccups.
- Build a fallback for AI dependencies. If Claude powers a production feature, decide in advance how you will pause jobs, queue requests, or switch models when it is overloaded.
- Turn on proactive alerting. Early Warning Signals exist for the window when something is clearly wrong but not yet officially confirmed.
- Track the provider’s reliability over time. Review Claude’s outage history and the Claude outage map to see status and outage data across dates and regions.
Frequently asked questions
Is Claude down right now?
This post covers the July 17, 2026 incident, which is resolved. For live status, check the Claude status page on StatusGator.
What is a 529 Overloaded error in Claude?
It is an HTTP response Claude returns when its servers are temporarily overloaded and cannot process the request. It usually clears on retry, but a sustained wave of 529s across many users signals a real outage.
How often does Claude go down?
Frequency varies. You can see the full record of tracked incidents on the Claude outage history page.
How can I get notified about Claude outages before they are announced?
Monitor Claude with StatusGator and enable Early Warning Signals, which alert you when user reports spike ahead of the official acknowledgment.
Monitor Claude with StatusGator
If you rely on Claude in production, you should not be the last to know when it is overloaded. StatusGator combines official status pages with real user reports and sends Early Warning Signals the moment a pattern emerges, often well before the provider confirms anything.
Start monitoring with StatusGator and get early warnings for the services your business depends on.
