Service Degradation Affecting Document Understanding, Agents, IXP, and GenAI Activities

Update timeline

1. investigating May 29, 2026, 04:18 PM UTC

   An issue with an upstream provider may be impacting GPT-4.0-powered functionality for Document Understanding, Agents, IXP, and GenAI Activities across multiple regions, and we are investigating. Impact: Users may experience request failures, increased error rates, or degraded performance when using GPT-4.0-powered features and workflows.

2. identified May 29, 2026, 05:03 PM UTC

   An issue with an upstream provider may be impacting GPT-4.0-powered functionality for Document Understanding, Agents, IXP, and GenAI Activities across multiple regions, and we are investigating. Impact: Users may experience request failures, increased error rates, or degraded performance when using GPT-4.0-powered features and workflows. Our teams have identified a potential mitigation and are actively implementing it to reduce customer impact. We will continue to provide updates as progress is made.

3. resolved May 29, 2026, 05:48 PM UTC

   The mitigation has been successfully implemented, and service has been restored. Impact has been resolved for Document Understanding, Agents, IXP, and GenAI Activities across all affected regions. Users should no longer experience request failures, elevated error rates, or degraded performance when using GPT-4.0-powered features and workflows. We will continue to monitor the services to ensure stability.

4. postmortem Jun 24, 2026, 07:34 AM UTC

   ## Customer impact Between May 29, 2026 at 2:40 pm UTC and May 29, 2026 at 5:48 pm UTC, a subset of customers experienced request failures, elevated error rates, and degraded performance when using GPT-4.0-powered features and workflows. The total customer-impacting duration was 3 hours and 8 minutes. Affected services included Document Understanding, Agents, intelligent extraction processing features, and generative AI activities. Impact was observed across multiple regions, primarily in Europe and Australia. ## Root cause The incident was caused by a service degradation at our upstream model provider, Microsoft Azure OpenAI Service. The affected model families — GPT-4.0, GPT-4.0 Mini, GPT-4.1, and text embeddings — returned failures, timeouts, and HTTP 5XX errors for requests routed to those deployments. Per Azure's Post Incident Review, the degradation was triggered by a provider-side change that caused a surge of internal retry traffic, which overwhelmed the shared routing layer Azure OpenAI uses to distribute inference requests. Because some UiPath services route to Azure OpenAI endpoints in the EU when a model is not available locally, the impact extended from the EU to dependent services in Australia. Azure has since mitigated the incident and confirmed remediation to prevent recurrence, including isolating large internal workloads onto dedicated infrastructure and strengthening overload controls. ## Detection The issue was first detected through automated alerting. Internal monitoring alert triggered on May 29, 2026 at 2:48 pm UTC — approximately 8 minutes after impact began \(the incident start was identified from logs as 2:40 pm UTC\) — and the team began investigating immediately. Shortly afterward, multiple other teams reported that they had received related alerts as well. In the window between this initial detection and formal incident coordination, we worked to characterize the issue, identify its likely cause, and determine the available mitigation options. At 3:43 pm UTC, we opened a dedicated incident channel and joined the response bridge to begin officially assessing customer impact and solutions. Follow-up log analysis confirmed elevated failures across the affected services and models. For context, Azure's monitoring first detected the broader incident at 9:39 am UTC during the earlier Australia East wave. UiPath's customer-facing impact and detection were tied to the later Sweden Central wave, which began at 2:40 pm UTC. ## Response * **2:48 pm UTC** — First alert from monitoring; Investigation began immediately. * **3:43 pm UTC** — Incident channel opened; formal impact and mitigation assessment began. * **3:49 pm UTC** — Affected regions identified as Europe and Australia. * **4:00 pm UTC** — Elevated failures confirmed across Document Understanding, Agents, and generative AI activities, with Document Understanding and Agents most affected. * **4:18 pm UTC** — Public status page updated to reflect degraded GPT-4.0 functionality across multiple regions. * **4:21 pm UTC** — Upstream provider \(Azure OpenAI\) identified as the cause. * **4:27 pm UTC** — Australia region recovered. * **5:09 pm UTC** — EU GPT-4.0 and GPT-4.0 Mini traffic redirected to an alternate model provider; verification confirmed recovery for the redirected models. * **5:24 pm UTC** — Remaining affected models recovered at the provider. * **5:48 pm UTC** — Mitigation confirmed complete and the incident resolved; service restored across all affected regions. ## Follow up **Alternate-provider routing readiness.** We have begun configuring the GPT model families currently served on Azure OpenAI to also be available via Native OpenAI for our EU and US regions, where Native OpenAI availability supports it. This establishes a fast, low-friction failover path: in a similar upstream Azure OpenAI incident affecting these regions, we can quickly redirect the affected GPT model traffic in the EU and US to Native OpenAI, reducing customer impact for those models and regions.