OpenAI's ChatGPT suffered a major global outage on April 20, with over 8,700 users reporting failures on Downdetector at its peak. The disruption lasted approximately three hours, taking down both ChatGPT and OpenAI's Codex coding tool simultaneously. OpenAI acknowledged the incident on its status page, stating it had "applied the mitigation and is monitoring the recovery."

The outage began around 10:05 AM ET on Sunday, April 20, when Downdetector reports spiked from a baseline of 13 to thousands within minutes. In the UK alone, reports peaked at 8,700. In the US, over 1,900 users reported failures. Of all complaints logged, 86 percent were tied specifically to the core ChatGPT service, with app issues at 8 percent and login failures at 4 percent.

Users attempting to access the platform via browser and mobile app alike encountered failed responses, loading errors, and complete inability to start conversations. The Downdetector graph showed a near-vertical climb in reports after 6 PM IST, with volumes remaining elevated for several hours before gradually tapering.

The dependency problem nobody wants to talk about

What makes this outage notable is not its scale but its timing. ChatGPT now has over 200 million weekly active users. Enterprises have embedded it into workflows ranging from customer service to code generation to legal document review. Codex, OpenAI's dedicated coding assistant, going down at the same time means developers relying on AI pair programming were left staring at blinking cursors.

This is the third significant ChatGPT outage in 2026. Each one follows the same pattern: sudden spike, hours of disruption, vague status page updates, eventual recovery. What never follows is an explanation of what went wrong or what changed to prevent a recurrence.

The incident comes at an awkward moment for OpenAI. The company is in the final stages of preparing for what would be one of the largest IPOs in history, with a valuation north of $850 billion. It is also in the middle of killing Sora (the video generation app shuts down April 26) and pivoting aggressively toward enterprise customers under CRO Fidji Simo's "kill the side quests" strategy.

A reliability problem at enterprise scale

Enterprise customers pay premium rates for reliability guarantees. When the core platform drops for three hours on a Sunday, the question every CTO asks on Monday morning is simple: what happens when this occurs on a Tuesday at 2 PM during a product launch?

OpenAI's competitors are watching. Anthropic's Claude has had its own reliability struggles, but the company has been transparent about capacity constraints. Google's Gemini operates on infrastructure Google has spent decades hardening for five-nines uptime. The reliability gap may not matter when AI tools are experimental. It matters enormously when they are load-bearing.

At the time of reporting, OpenAI had not issued a formal public statement beyond its status page update. The company historically communicates through status.openai.com during incidents but rarely provides post-mortems. For a company positioning itself as the operating system of enterprise AI, silence after outages is not a strategy. It is a liability.