ChatGPT Goes Dark Worldwide — OpenAI Confirms Major Outage Crippling API, Codex, and Hundreds of Third-Party Services
On the morning of July 25, 2026, ChatGPT stopped working on a planetary scale. Users across the United States, Europe, India, Japan, and Australia were met with error screens where instant responses had lived moments before. But the disruption went far beyond the chatbot. OpenAI confirmed through its status page that the incident simultaneously crippled ChatGPT, its developer API, and Codex — the company's coding assistant. In total, 15 ChatGPT components, 12 API components, and 4 Codex components were flagged as affected, amounting to a near-total paralysis of the company's ecosystem.
A rough week
The July 25 blackout was not an isolated event. On Thursday, July 23, around 11:30 AM ET, OpenAI had already logged elevated error rates across the same services. The University of Colorado's IT office, which mirrors OpenAI incidents for its own users, attributed the cause to "a downstream infrastructure provider" — a problem in a third-party layer that OpenAI depends on. A mitigation was applied hours later, but by the evening a second wave hit ChatGPT, lasting nearly three hours. Independent tracker StatusGator computed roughly eight hours and twenty minutes of degraded service on July 23 alone.
By Saturday, July 25, the pattern repeated with greater force. Starting around 5:00 AM ET, thousands of reports flooded Downdetector, the world's most used outage aggregator. Users encountered 503 errors carrying the internal codename "biscuit_baker_service_me_circuit_open" — an cryptic message indicating that requests weren't even reaching OpenAI's servers. For many subscribers, chat history — one of ChatGPT's most valued features — simply vanished.
Real impact: far beyond broken chat
When ChatGPT went down in 2024, the impact was limited to lost conversations and some social media frustration. But the OpenAI of mid-2026 is a different company. In May, under Greg Brockman's leadership, the company unified ChatGPT, Codex, and the API into a single platform — precisely to concentrate engineering resources. The result is that when something breaks, everything breaks together.
The company now serves 900 million weekly active users. Of that total, 10 million are "agentic" users — autonomous systems that execute multi-step tasks on behalf of humans, from scheduling meetings to managing data pipelines. When an autonomous agent goes down mid-task, the consequences cascade through every workflow it was managing. These aren't just lost prompts: they're missed deadlines, broken integrations, and disrupted business operations.
ChatGPT Work, launched this very month of July, promised exactly this: agents capable of crossing applications, files, and knowledge bases to complete complex tasks. The promise depends entirely on availability. Each outage erodes the confidence that the platform is ready to take on critical responsibilities.
A troubling pattern
OpenAI's status history reveals a concerning trend. Since fall 2025, the status page has logged approximately 166 incidents over nine months — an average of 18 per month. In April 2026, a major disruption took ChatGPT offline. In July alone, image generation went down on the 21st, a 45-minute outage hit on the 15th, and another 44-minute outage struck on the 7th. The cadence suggests infrastructure is not keeping pace with adoption.
OpenAI is not alone in this struggle. Anthropic's Claude suffered a significant outage in June 2026, raising the same questions about whether the infrastructure behind frontier AI models can sustain the demand they generate. The difference is one of scale and ambition: OpenAI is preparing for an IPO later this year, and every outage chips away at the reliability narrative enterprise buyers need before handing real work to an AI agent.
What comes next
For businesses already integrating OpenAI's API into their products, the lesson is clear: depending on a single provider is risky. The multi-provider fallback strategy, once seen as theoretical precaution, is becoming an operational necessity. If one provider goes down, the system needs to reroute — and fast.
The open question is whether AI-as-a-service infrastructure will mature in time to sustain the promises the industry is making, or whether the wave of adoption will break first on the shore of reliability.
Sources: BleepingComputer, The CyberSec Guru, Sunday Guardian Live, TNW (The Next Web), Unite.AI
