GitHub Outage Disrupts Copilot and API Services

GitHub experienced a widespread service disruption on August 17, 2026, affecting its website, API, Pull Requests, Webhooks, and the AI tool Copilot. The outage began around 9:40 a.m. EDT and lasted until 21:15 UTC, with error rates reaching 20% for web and API traffic and 50% for archive and raw repository content downloads. Enterprise identity services, including SAML and OIDC authentication, were also impacted.

GitHub attributed the failure to network saturation on load balancers in its Central US datacenter. The situation was worsened by a misconfigured autoscaling policy for an Istio sidecar pod and a latent retry bug in VS Code that increased Copilot Token Service traffic by approximately 10x. Recovery efforts were further hindered by scraping attacks on codeload endpoints and a retry storm in Northern Virginia after traffic was rerouted.

Engineers stabilized the system by pausing HAProxy nodes, reducing gateway retry logic, and blocking specific token requests. While services returned to normal by late afternoon, the incident follows a period of decreased reliability with 14 logged incidents in August. Microsoft, the parent company of GitHub, also saw minor reports of issues with Microsoft Teams and Copilot during the same window.