What Went Wrong

ChatGPT went down for users around the world on Saturday, July 25, 2026, marking the fourth service disruption for OpenAI in as many days and leaving free, Plus, and Enterprise users alike unable to access the platform.

The outage began in the early morning hours, around 5 a.m. ET, with users reporting they were unable to load previous conversations, send new messages, or even get past loading animations in the sidebar. Many encountered “too many concurrent requests” errors, while others saw their screens stuck in endless loading loops.

OpenAI’s status page acknowledged the problem shortly after 5:30 a.m. ET, stating simply that the company was “investigating the issue for the listed services” — naming its API, ChatGPT, and the Codex coding assistant as all affected simultaneously.

Later reporting indicated users encountered 503 errors carrying the internal label “biscuit_baker_service_me_circuit_open,” which blocked requests from reaching OpenAI’s servers entirely. The company reportedly moved from “investigating” to “monitoring” within about an hour after applying a mitigation, though full recovery took longer to confirm.

How Widespread Was It

The disruption wasn’t limited to the United States. Reports poured in from users in Europe, India, Japan, and Australia, according to outage-tracking sites. Downdetector logged a sharp spike in complaints, jumping from a typical baseline of just a couple of reports to more than 1,500 in a short window, with ChatGPT itself accounting for roughly 80% of the flagged issues.

Some early analysis pointed to the login and authentication layer as the likely source of the problem rather than the underlying AI models themselves, though OpenAI did not confirm a root cause in its public updates.

Who Was Affected

• Free, Plus, and Enterprise users on both web and mobile apps reported being locked out or stuck in loading loops.
• Developers and businesses relying on OpenAI’s API for their own applications saw their integrations fail alongside the consumer product.
• Codex users working on coding tasks were also cut off, with many taking to social media to vent about disrupted workflows, writing projects, and coding sessions.

Not a One-Off Incident

This marks the fourth outage OpenAI has dealt with in just four days, according to the company’s own status page, raising fresh questions about the reliability of its infrastructure as ChatGPT usage continues to scale globally. Outage-tracking service IsDown, which has monitored the platform since October 2025, notes that OpenAI incidents have historically taken an average of around five hours to resolve.

The Bigger Picture

ChatGPT has weathered high-profile outages before, including a lengthy disruption in June 2025 that kept the service down for more than ten hours. As AI chatbots become embedded in daily work for millions of people — from coding and writing to customer service — even brief outages now ripple across businesses that have built their tools on top of OpenAI’s platform.

OpenAI has not yet issued a detailed post-incident report explaining the underlying cause of Saturday’s disruption or outlining steps to prevent future recurrences.

This is a developing story and will be updated as OpenAI releases more information.