What happened
Claude, Anthropic’s AI service, had a major outage on July 21, 2026. It started at 17:41 UTC and recovered at 18:04 UTC, lasting about 23 minutes. Anthropic has marked the issue resolved. No detailed cause has been published at this time.
Who this kind of outage hits, and how they usually find out
If you have built anything on Claude, whether that is a workflow that drafts content, a support tool that reads customer messages, or an app that calls the API directly, a 23-minute outage means every one of those things stopped producing results. The work queued up silently. Nothing on your screen turned red. The most common way operators find out is a customer writing in to say something is broken, or a colleague asking why a report did not arrive. By the time that message lands, the outage may already be over, but the damage to trust is done.
Why this is especially rough if you are not an engineer
There are no logs to pull up. There is no error message sitting somewhere obvious. The work just stops moving, and the silence looks exactly like normal. A non-engineer operator has no way to tell the difference between “Claude is down” and “something I did broke my workflow” without spending time they do not have digging through settings. The first real signal is often an unhappy customer, and that is a terrible way to learn about a third-party outage that was never your fault.
Timeline
- 17:41 UTC - Claude stops working. Major outage begins.
- 18:04 UTC - Service recovers. Total disruption: about 23 minutes.
How a watcher catches this before your users do
Claude publishes a public status page. NoCrash reads that page every minute. The moment it flips from working to having trouble, NoCrash sends you a plain-language message telling you what broke and when. You do not get a raw status code or a vague alert. You get something a non-engineer can read and act on immediately, sitting next to the status of everything else you build on.
NoCrash also watches the things you ship. If you run n8n workflows, it watches those through your API token. If you have an app, it watches it through a URL you give it or a small JS snippet you drop in. So if Claude goes quiet and your workflow stalls as a result, you see both signals in one place.
The honest limit: NoCrash reads Claude’s own public status page. It catches the trouble within a minute of Anthropic’s own report. It does not find the outage before Anthropic does. What it removes is the gap between Anthropic knowing and you knowing, which is usually where the “my customer told me first” problem lives.
The authoritative account
For the official record of this outage, see Anthropic’s status page: https://stspg.io/4lm8kwq5bpfy