A large-scale digital outage impacted artificial intelligence platforms, banks and several government services in Brazil on September 3, 2026. AI tools such as ChatGPT, Claude and Grok became inaccessible almost simultaneously, triggering a ripple effect across several online operations.
The simultaneous incident generated concern in the technology market. In the Brazilian scenario, monitoring by DownDetector, a platform that tracks online interruptions, pointed out instabilities in essential services such as Gov.br, Banco do Brasil and Caixa Econômica Federal in the same period of the outage.
Platforms and systems affected by digital instability
The strike affected operational routines, programming projects and access to services considered essential. The situation led to the investigation of the following facts:
- Dead AI robots: The developer companies themselves confirmed the crashes of ChatGPT, Claude and Grok.
- Unstable Brazilian services: DownDetector recorded a significant increase in complaints about Gov.br, Caixa and Banco do Brasil.
- Code tool stalled: Agent Cursor, who depends on Claude and Grok to operate, has also stopped working.
- Platform in operation: Google Gemini did not experience any failures, operating normally during the period of the problem.
Suspected cloud infrastructure failure
The source of the problem appears to converge on the infrastructure that hosts these services. The three major artificial intelligence platforms affected use Microsoft Azure servers to maintain their operations. Despite the existence of other cloud services involved, Azure is a common link.
With a significant increase in complaints registered with Azure in the same time period, the main technical hypothesis points to a failure in the software giant’s data centers. The centralization of large services in a few cloud infrastructures can generate broad impacts, affecting several interdependent systems in the event of a single failure.
Cloudflare, a company that has been linked to other major recent cyber outages, does not appear to be involved in the incident. The company’s chief technology officer, Dane Knecht, stated that Cloudflare’s services are not currently facing disruption.
Efforts to normalize services
Engineering teams are actively working to stabilize connections and restore affected servers to full operational capacity. Users can monitor the recovery of each tool through the status panels provided by the respective companies.
On DownDetector, the number of issue notifications began to decrease rapidly in the afternoon of September 3, 2026. The status pages for Claude and ChatGPT already indicated that all services were operational in the early afternoon, while Grok was still experiencing some errors.
