What happened
On the evening of August 14, 2026, Claude had a bigger problem that lasted about two hours. It started at 20:31 UTC and recovered at 22:48 UTC. Anthropic has since marked it resolved. That is the full picture the official record gives us right now.
Who this hits and how they usually find out
If you build on Claude, your automations, your writing tools, your support flows, anything that calls the API quietly stopped getting useful responses during that window. The hard part is that nothing inside your product screams at you. There is no red light. The work just stops moving. Most operators find out the same way: a customer writes in to say something is broken, or they notice a backlog of tasks that never ran. By then the outage is already an hour old, sometimes more. The gap between “it broke” and “I knew it broke” is where the damage happens.
Why this is especially rough without a technical background
If you are not an engineer, you have no logs to pull, no error trace to read. You just see that something did not happen that was supposed to happen. A workflow that should have sent a message did not. A summary that should have appeared is blank. The first signal is almost always a confused or unhappy customer, not a system alert. That is a terrible position to be in, because by the time you are explaining it to them, you are already behind.
Timeline
- 20:31 UTC Claude started having a bigger problem.
- The disruption ran for about two hours with no recovery in between.
- 22:48 UTC Claude recovered and Anthropic marked the issue resolved.
How a watcher catches this before your users do
NoCrash reads Claude’s official public status page every minute. The moment that page flips from working to having trouble, NoCrash sends you a plain-language message telling you what is wrong, in words you can act on, without you having to go check anything yourself. That alone turns “my customer told me two hours later” into “I got a heads-up within a minute of Anthropic’s own report.”
It also watches the things you ship. If you have n8n workflows, NoCrash watches those through your API token. If you have an app, it watches that through a URL you give it or a small JS snippet you drop in. So if the trouble is on your side, a quiet stall in your own workflow surfaces the same way, in the same place, without you having to piece it together from two different tools.
To be clear about what this is: NoCrash reads what Anthropic publishes. It does not find the outage before Anthropic reports it. What it does is make sure you are not the last to know once Anthropic does report it.
The authoritative account
For the official record of this outage, go to Anthropic’s status page: https://stspg.io/wlcwm73lh5vb