7:25 AM: "Claude Won't Open"
On the morning of August 17, 2026, Claude stopped working worldwide at the same time.
A post from the news account The Spectator Index reading "BREAKING: Claude outage for many users across the globe" drew more than 380,000 views.
Reactions piled up: logins failing, answers cutting off, and screens freezing in the middle of coding work.
42 Minutes From Authentication Errors to Degraded Performance
According to Anthropic's official status page, the outage began at 21:58 UTC on August 16, 2026 (6:58 AM KST on August 17).
The first notice was "authentication issues for some users of claude.ai, Claude Code, and Claude Cowork."
Four minutes later, at 22:02 UTC (07:02 KST), Anthropic widened the scope and officially acknowledged "degraded performance on claude.ai and platform.claude.com."
A fix was deployed at 22:22 UTC (07:22 KST), and all services were confirmed back to normal between 22:34 and 22:40 UTC (07:34 to 07:40 KST). The total outage lasted about 36 to 42 minutes.
Anthropic did not disclose the specific cause.
| Time (KST) | Event |
|---|---|
| 06:58 | Outage begins: notice of investigation into authentication errors on claude.ai, Claude Code, and Claude Cowork |
| 07:02 | Scope widens: degraded performance on claude.ai and platform.claude.com officially acknowledged |
| 07:25 | The Spectator Index "global outage" post spreads (380,000+ views) |
| 07:22 | Fix deployed |
| 07:34 to 07:40 | All services confirmed back to normal |
Why Does a Different Account Always Post the "Breaking News" First?
As with this outage, when Claude goes down, X posts from news accounts and individuals spread before the official status page does.
In fact, "Claude is down" breaking-news posts went up from multiple accounts at the same time during this outage.
That's because Anthropic's status page (status.claude.com) updates its notices a few minutes late, while users feel the problem instantly.
This isn't the first major Claude outage. In March 2026, there was a separate outage with about 2,000 simultaneous connection error reports on Downdetector.
What You Can Do Now to Prepare
- If you depend on Claude Code or Cowork for work, bookmark status.claude.com so you can find the cause faster.
- If your product integrates the API, building in short retries and timeout handling by default reduces the outage users feel in moments like this.
- Outages like this one often end within 30 to 40 minutes, but since the cause hasn't been disclosed, checking the status page yourself is the most reliable way to know whether it recurs.
Anthropic hasn't separately disclosed the root cause or the steps to prevent a recurrence, so these are not officially confirmed.