ChatGPT and Codex Hit Global Outage as OpenAI Investigates

OpenAI is in the middle of a global outage affecting ChatGPT, Codex, and the API platform. The problems started around 10:05 AM ET on Monday, April 20, and thousands of users across the US, Europe, and Asia are reporting they can't load conversations, log in, or generate images. OpenAI's status page has upgraded the incident from "investigating" to "partial outage" and says mitigation is underway.

This is the second major ChatGPT outage of 2026 and the most widespread so far. For a product that now serves roughly 900 million weekly active users — many of them running production business workflows — even partial downtime cascades quickly.

Why this matters: Enterprise dependence on a small number of frontier AI providers means that a single OpenAI incident can freeze coding teams, stall customer support pipelines, and halt agent workflows at thousands of companies simultaneously. The outage is a live demonstration of concentration risk in the AI stack.

What's Affected

OpenAI's status page lists multiple affected components. Users are reporting they can't load ChatGPT conversations, can't sign in to their accounts, can't use voice mode, and can't generate images through DALL-E integration. The Codex coding platform is showing degraded performance. API customers are seeing elevated error rates on multiple endpoints.

Not every user is seeing identical symptoms. Some can load ChatGPT but fail at message sending. Others are blocked at login. Still others have partial functionality — text works, but image generation fails. The uneven pattern suggests multiple downstream services are affected by an upstream issue OpenAI hasn't fully diagnosed.

Downdetector reported a spike of thousands of incident reports within the first hour, with concentrations in North America and Europe. By early afternoon ET, the report volume hadn't dropped — meaning recovery is partial at best.

What OpenAI Has Said

The company's communications have been minimal and status-page-only. OpenAI acknowledged it was "investigating users unable to load ChatGPT and Codex" and broader "degraded performance." A later update confirmed the company had "applied mitigation and is monitoring the recovery."

OpenAI has not disclosed a root cause, an estimated time to full resolution, or whether the incident is related to recent infrastructure changes. The company's status page is the primary official communication channel; there has been no blog post, social media update, or press statement beyond that.

This pattern — minimal communication during active incidents, fuller post-mortems days later — is standard across major cloud providers. It's also why third-party status trackers like Downdetector and StatusGator get so much traffic during outages: official sources often lag user-reported reality by 30 minutes or more.

Why Production Teams Are Watching

For consumer users, an OpenAI outage is annoying. For enterprise customers, it's expensive. Many companies have built workflows that assume ChatGPT or the OpenAI API as a hard dependency — customer service tooling, code review pipelines, research assistants, agent systems that make autonomous API calls.

When OpenAI goes down, those workflows either fail or silently fall back to older, less capable models if the engineering team built in fallback logic. Most teams haven't. The result is that a single-provider outage can ripple into real revenue impact within minutes.

The problem compounds at the infrastructure layer. OpenAI's API is behind a huge number of downstream products — including applications from companies that never disclose their provider to end users. When ChatGPT has a bad hour, dozens of B2B SaaS products have a bad hour simultaneously, often without their customers understanding why.

Multi-Provider Strategies Are No Longer Optional

Enterprise AI buyers have been talking about multi-provider strategies for over a year. Monday's outage is a reminder that it's no longer a theoretical best practice. Teams running meaningful production traffic on OpenAI should have Anthropic's Claude API, Google's Gemini API, or an open-source alternative wired in as a fallback — ideally with automatic failover.

The frameworks exist. LiteLLM, OpenRouter, and several orchestration platforms offer drop-in abstraction layers that route requests to whichever provider is available. The marginal engineering cost is real but bounded. The cost of not doing it is what thousands of teams are experiencing right now.

The irony is that multi-provider setups have gotten easier, not harder, over the past year. Claude, Gemini, and GPT have converged on similar API shapes. Prompt portability has improved. Capability gaps between the top three labs are narrow enough that a well-designed fallback often degrades gracefully rather than failing visibly.

What to Watch Next

OpenAI will eventually publish a post-mortem, likely within a week. Key questions: Was this a regional issue that cascaded globally? An infrastructure change gone wrong? A DDoS or security incident? The answer shapes how customers should adjust their reliability planning.

For users, the practical response right now is patience. OpenAI says mitigation is underway, and historical precedent suggests full recovery within 2-6 hours of the initial incident start. For enterprise teams with automated systems still failing, this is the moment to check whether your fallback logic actually works — or whether you've been assuming it did.

The Bottom Line

ChatGPT and Codex are partially down globally as of Monday afternoon. OpenAI says it's working on it. If you rely on the OpenAI API in production, today is a good day to finally wire up that multi-provider fallback you've been meaning to build. And if you're just trying to finish a task, try again in a few hours — recovery is typically measured in hours, not days.