It didn’t take long for Anthropic to put things right. But the fact that they had to restore access after a disruption to so many of their services is a sign of how much usage has been on the upswing. With thousands of reports filed with Downdetector, it’s clear just how much teams and developers have come to rely on the AI for the day-to-day.
Here is what you need to know from the incident:
– We saw reports top 2,500 in the 12:33 PM hour
– 2,174 issues were put in the book by Downdetector at 12:34 pm
– By 12:43 or so, the number was over 2,200
What happened and how it unfolded
The numbers on Downdetector’s graph don’t tell a story of a slow build-up; the spike was sudden. It came on the heels of a fairly uneventful 24 hours, which points to a hard stop in service as opposed to some kind of ongoing wobble.
Anthropic put out a word on its status page to the effect that they had found the problem and were on it. Services were back up by 1:48 PM IST. The official window for the trouble was 06:55 to 07:40 UTC – not a long time, but it was felt in more than one place on the Claude side of things.
If you look at the various snapshots of the event, the scale is there. Downdetector has the data: 2,500-something reports at 12:33 PM, 2,174 at 12:34, and then over 2,200 a few minutes later.
User impact and what people saw
For the most part, it was a matter of prompts not going through, empty responses and the like. Downdetector says 60% of the noise was about Claude Chat. You could also make a case for 63% on the bot, 25% on the app and 10% on the site if you read it another way.
There were plenty of "can't load” and “won’t let me in” type of comments. One dev had a laugh about being told to code without Claude, while someone else was wondering if they were the only one who was out. Then you had the user in the thick of a job when the assistant just went quiet.
One outlet put it to the test and, after a bit of a wait, the bot did put together some code for a coffee shop site. So it was coming back online before the all-clear was given. Anthropic didn’t have a comment to make in the moment, but they did confirm it was over on their end.
Strategic stakes for Anthropic
This isn’t the first time we’ve heard of a hiccough with Claude this month, and it’s part of a run of them in 2026. If you check the logs for June, you’ll see some higher error rates and performance matters with both the .ai and the API.
We’ve seen some of this with Opus 4.8 and Haiku 4.5 as well, where things would slow down or flat out fail. Anthropic will be the first to tell you that the appetite for their tools, from Claude Code to the enterprise side, has been hard to keep up with.
They are in the process of putting more compute in place with some of the big cloud names, even if not all of it is ready to go yet. On top of that, a US government order on national security has put a hold on Fable 5 and Mythos 5 for anyone outside the country.
Why it matters and what comes next
When you’re using an AI to write, do research or code, 45 minutes without it can throw a wrench in your schedule. This is why being able to scale is the name of the game for these platforms now.
Anthropic is on top of it and says normal is returning. If you are having any trouble, the status page is where you want to be. And while Downdetector’s stats are only as good as the users who put them in, they are a fair barometer of what’s going on.
