Simultaneous outage across major chatbots: ChatGPT down for about two hours

This morning’s edition reported on the release of GPT-6 Astra and the limits OpenAI says it has imposed on its cybersecurity functions. This afternoon’s new development is a different one, concerning infrastructure rather than the models: on that same day, September 3, the main conversational assistants ran into trouble almost simultaneously.

Timeline of the outage

Problems on ChatGPT began around 4:30 PM on September 3 and largely subsided by around 6:00 PM, according to the reconstruction by Fanpage.it, which is based on OpenAI’s own status page and on reports collected by Downdetector. The status.openai.com page recorded a high number of errors on ChatGPT and Codex, affecting conversations, account access, file uploads and image generation: not a single component, then, but several functions at once.

Napolike quantifies the outage at around two hours and more than 38,000 reports collected worldwide. The number of reports measures users’ reaction, not the number of people actually affected: it is an indicator of perceived intensity, and should be read as such.

A third element comes from ZetaNews: OpenAI officially acknowledged an incident in the early hours of September 3, concerning ChatGPT Work Mode, with error rates above normal, later declared resolved. These are therefore two distinct episodes on the same day: one acknowledged by the company in the morning, one recorded by external monitoring in the afternoon.

Not just one company

What makes the episode notable is the simultaneity. Malfunctions were also reported on Anthropic’s Claude, Google’s Gemini and, to a lesser extent, xAI’s Grok. Three different companies, with different computing infrastructures, running into problems within the same time window.

In the excerpts available to us there is no technical explanation for this coincidence, nor a common cause indicated by anyone: none of the companies involved, as far as can be established, has linked its own disruptions to those of the others. This newspaper records the simultaneity as a fact and stops there, because the next step — hypothesizing a shared provider, a common network service or a single origin — would be a deduction, not a piece of news.

The coincidence with the release

September 3 is also the day OpenAI released GPT-6 Astra. The company’s president, Greg Brockman, called it a generational leap and said he personally believes the company has achieved artificial general intelligence; this is reported by laRegione, based on a meeting with the media, and by Quotidiano Nazionale.

“I believe this model may represent the milestone” — Greg Brockman, president of OpenAI

Here too, no element in the available sources links the release of the new model to the disruptions of the following hours. It should be noted, however, that among the services with errors was Codex, the programming tool the company updated alongside Astra, for which it claims a 1.9-fold speed improvement over the previous model on the Mind2Web test — a figure stated by the company and not independently verified.

What we do not know

We do not know how many people were actually affected by the outage, nor which geographic areas. We do not know the technical cause stated by OpenAI for the afternoon episode, nor whether the other companies have published their own account. We do not know whether customers paying for a professional subscription received any information about the service.

The verifiable fact is the duration: about two hours of degraded operation on tools that millions of people now use as part of their daily workflow.