The five big chatbots in the AI universe have stopped working at the same time. OpenAI's conversational assistant, ChatGPT; that of X, Grok; Anthropic's, Claude; Microsoft's, Copilot; and Google's, Gemini, have caused problems for their users around the world starting at three in the afternoon, Spanish peninsular time. The Down Detector service, which monitors the fall of internet services, has reported problems in the most popular artificial intelligences. Users have reported problems preventing the use of both the website and the mobile application.
Although the first to fail were Grok and Claude, around 3:00 p.m., later, around 4:30 p.m., ChatGPT, Gemini and Copilot joined. As of 6:30 p.m., reports of users complaining about the service failure had significantly reduced, although Grok, Claude and Copilot continued to cause problems.
"ChatGPT is experiencing a high error rate. We are investigating the problem," the OpenAI company stated on the website they have to report on the operation of the service.
A similar message was posted by Claude, who has determined that the problems have lasted 3 hours and 6 minutes: "An infrastructure problem is causing a partial crash in Claude Code, the API and Claude. We are working on it," wrote Anthropic engineer CJ Avilla in X. The company has not given more details, but that reference to "infrastructure" points to a problem outside the company.
In October and November 2025, two massive outages left a lot of essential pages on the network without service, including several AI assistants. So, the cause was Amazon Web Services first and Cloudlflare, two of the pillars of hosting key pages on the Internet. Today there have also been reports of problems with these two companies and Microsoft Azure, which also hosts cloud services. The problem has been, however, clearly smaller than then, due to the duration and number of services affected. Cloudflare's chief technology officer, Dane Knecht, ran to
ChatGPT or Claude are services mounted on infrastructure that belong to these other companies: Amazon (AWS), Microsoft (Azure) or Google manage the servers where these models are located, and Cloudflare acts as an intermediary between users and those computers. It's a chain. If a link breaks, everything can stop working well. Thus, a Cloudflare issue causes a query to not even reach its destination even though Grok or Gemini are working perfectly. If Azure or AWS fail, the model cannot run directly.
All of these companies rent the same infrastructure. Therefore, a simple failure can seem like it brings about the end of the world and brings down services that actually compete with each other and are independent. Thus, a problem in a provider that few people know about can leave half the Internet, in this case several AI bots, out of work for a while.
This fall also occurs at a time when AI seems to be taking off towards a world of science fiction: since mid-July, several companies have announced that agents or autonomous systems driven by their models had hacked other companies, created communities with message exchange and sacrificed robots, or have written to humans to trick them and achieve their goals. The distance between those feats and the decline of these basic services may seem surprising, but it is the daily life of the Internet today. Not all companies do the same.
This Thursday, in addition, OpenAI has hinted with a message also in X that its new family of models, Astra, is about to arrive. The company said it was its first model to cross the critical threshold of capability in cybersecurity because it can find unknown security flaws and exploit them without human involvement. For this reason, it was decided to delay its launch and the company points out that it will limit access to these capabilities. The majority of responses on X to that trailer were along the lines of: “Yes, but ChatGPT is down now.”
