The latency event was reported by multiple outlets. 9to5Mac noted that by 8:15 a.m. PT, a separate FedRAMP issue was also being monitored, though a fix had already been applied. The ChatGPT latency was reportedly resolved first, with API performance recovering fully shortly after . Android Authority confirmed the resolution at 4:28 p.m. ET, noting that the incident had affected users throughout the morning . OpenAI did not publicly provide a detailed root cause.

May 26–27: FedRAMP Users Locked Out for 23 Hours

A separate and more prolonged incident began on May 26, specifically affecting FedRAMP-authorized users—the U.S. government's secure cloud program. Users who had recently logged out found themselves unable to log back into ChatGPT, effectively blocking access to the service .

The outage lasted approximately 23.2 hours before OpenAI applied a mitigation and began monitoring recovery . OpenAI's own status page confirmed the fix and ongoing monitoring, while third-party monitors like EagleStatus and Pagerly listed the FedRAMP login issue as ongoing through May 27 . 9to5Mac later reported that the FedRAMP incident had been removed from the status page after the fix was fully confirmed .

The long duration of this incident is notable given that FedRAMP is used by government agencies that often require higher reliability standards.

A Pattern of Minor Outages Across May 2026

Beyond the two high-visibility incidents in late May, OpenAI experienced at least six other minor disruptions throughout the month, primarily affecting the GPT-5.5 and GPT-5.4 model families, as well as the Codex developer tooling. All were resolved, and none were classified as major outages. The table below summarizes each event based on official status-page data and third-party incident trackers.

What This Cluster Means

Although each incident on its own was minor, the concentration of issues around GPT-5.5, GPT-5.4, ChatGPT 5.5 Thinking, and Codex points to teething problems for OpenAI's newest generation of models . The GPT-5.5 family had only fully launched to paid ChatGPT users in late April 2026, with API access following on April 24 .

The May 15–16 performance degradation was the longest individual event, persisting for roughly 32 hours and affecting GPT-5.5 performance without completely breaking the service . The May 22 Codex rate-limit spike, lasting over 18 hours, appeared to affect a higher-than-normal number of users trying to access coding features, potentially indicating capacity constraints or configuration issues after recent product updates .

OpenAI’s status page shows that the API layer maintained a 99.98% uptime over the 90-day period ending in May 2026, with the broader system at 99.84% . These numbers are consistent with a platform that is generally healthy but experiencing recurring, short-lived instability on specific cutting-edge components. For teams relying on the latest models for production workloads, May 2026 served as a reminder that even minor-seeming incidents can disrupt workflows when they arrive in tight clusters.

The latency event was reported by multiple outlets. 9to5Mac noted that by 8:15 a.m. PT, a separate FedRAMP issue was also being monitored, though a fix had already been applied. The ChatGPT latency was reportedly resolved first, with API performance recovering fully shortly after . Android Authority confirmed the resolution at 4:28 p.m. ET, noting that the incident had affected users throughout the morning . OpenAI did not publicly provide a detailed root cause.

May 26–27: FedRAMP Users Locked Out for 23 Hours

A separate and more prolonged incident began on May 26, specifically affecting FedRAMP-authorized users—the U.S. government's secure cloud program. Users who had recently logged out found themselves unable to log back into ChatGPT, effectively blocking access to the service .

The outage lasted approximately 23.2 hours before OpenAI applied a mitigation and began monitoring recovery . OpenAI's own status page confirmed the fix and ongoing monitoring, while third-party monitors like EagleStatus and Pagerly listed the FedRAMP login issue as ongoing through May 27 . 9to5Mac later reported that the FedRAMP incident had been removed from the status page after the fix was fully confirmed .

The long duration of this incident is notable given that FedRAMP is used by government agencies that often require higher reliability standards.

A Pattern of Minor Outages Across May 2026

Beyond the two high-visibility incidents in late May, OpenAI experienced at least six other minor disruptions throughout the month, primarily affecting the GPT-5.5 and GPT-5.4 model families, as well as the Codex developer tooling. All were resolved, and none were classified as major outages. The table below summarizes each event based on official status-page data and third-party incident trackers.

What This Cluster Means

Although each incident on its own was minor, the concentration of issues around GPT-5.5, GPT-5.4, ChatGPT 5.5 Thinking, and Codex points to teething problems for OpenAI's newest generation of models . The GPT-5.5 family had only fully launched to paid ChatGPT users in late April 2026, with API access following on April 24 .

The May 15–16 performance degradation was the longest individual event, persisting for roughly 32 hours and affecting GPT-5.5 performance without completely breaking the service . The May 22 Codex rate-limit spike, lasting over 18 hours, appeared to affect a higher-than-normal number of users trying to access coding features, potentially indicating capacity constraints or configuration issues after recent product updates .

OpenAI’s status page shows that the API layer maintained a 99.98% uptime over the 90-day period ending in May 2026, with the broader system at 99.84% . These numbers are consistent with a platform that is generally healthy but experiencing recurring, short-lived instability on specific cutting-edge components. For teams relying on the latest models for production workloads, May 2026 served as a reminder that even minor-seeming incidents can disrupt workflows when they arrive in tight clusters.