Thousands of users faced disruption on Tuesday, February 3, 2026, as ChatGPT went offline in a significant outage that affected access across the United States and international regions. The popular AI chatbot, developed by OpenAI, stopped responding to queries for many subscribers, leading to a rapid spike in error reports on outage tracking platforms. OpenAI has since confirmed that the incident is resolved, having identified the underlying issues and deployed a fix to bring services back online.
Reports Surge into the Thousands
The disruption began midday Pacific Standard Time, with users immediately taking to social media and status trackers to report their inability to access the service. Data from Downdetector, a platform that aggregates status reports from various sources, illustrated the scale of the problem. By 12:46 p.m. PST, more than 20,000 users had already flagged issues with the platform.
The volume of complaints continued to grow rapidly over the next hour. Just five minutes later, at 12:51 p.m. PST, the number of reports had climbed to over 22,000. The surge showed no signs of slowing down immediately, as reports surpassed 25,000 by 1:03 p.m. PST. At the height of the incident around 1:08 p.m. PST, Downdetector data indicated that more than 28,000 users were experiencing problems with the AI tool.
While the majority of these reports appeared to originate from the United States, users in other countries also faced difficulties. Observations noted that the service was down in Hungary and the United Kingdom, suggesting the technical glitches were not isolated to a single region.
OpenAI Confirms Two Distinct Issues
OpenAI acknowledged the disruption on its official status page shortly after the problems began. The company pinpointed the start of the issues to approximately 12:32 p.m. PST. In their updates to users, OpenAI specified that they were dealing with two separate but simultaneous technical complications.
The first issue involved “elevated error rates for finetuning jobs,” affecting developers and power users attempting to customize models. The second, more widespread problem was described as “elevated errors for ChatGPT and Platform users.” This latter issue was responsible for preventing the general public from sending messages or receiving responses from the chatbot.
During the outage, the company’s status messages were transparent about the ongoing investigation. They stated, “We have identified that users are experiencing elevated errors for the impacted services. We are working on implementing a mitigation.” This confirmation helped validate the frustration of thousands of subscribers who found themselves staring at blank screens or error messages instead of the usual conversational interface.
User Experience and Workarounds
For those trying to use the tool during the downtime, the experience varied. Many users on the free tier found the service completely unresponsive. In some cases, prompts would simply hang without generating a reply, or the interface would take significantly longer than usual to process a request.
Interestingly, the impact appeared to be uneven across different network configurations. Reports indicated that while users in the UK were unable to get a response, some found success by using a Virtual Private Network (VPN) to route their connection through the United States. This workaround suggested that the mitigation measures or the errors themselves might have been propagating differently across global servers.
Even some premium subscribers were not entirely immune, though their experiences were mixed. Some users with ChatGPT Plus accounts reported being able to log in and receive answers, although the system was noticeably slower. Others shared screen grabs of the interface failing to respond altogether, regardless of their subscription status.
Services Restored and Mitigations Applied
The outage, while widespread, was relatively brief in duration. By late afternoon, the situation had improved significantly. OpenAI updated its status page to inform users that they had successfully identified the root cause and applied necessary mitigations. The company noted that they were “monitoring the recovery” to ensure stability returned to the platform.
External data supported this timeline of recovery. The number of reports on Downdetector dropped sharply from the peak of 28,000 down to under 300 in the United States by 4:28 p.m. ET, indicating that the vast majority of users were back online.
Upon returning to functionality, the chatbot itself seemed to be working normally. When asked about its status following the fix, one instance of the AI described itself as “a little rebooted” and “feeling pretty good,” confirming that the system was once again processing queries with its characteristic conversational style. With the fix in place, operations have returned to normal, ending a brief but disruptive afternoon for the AI community.
