GitHub’s August 17 Outage Disrupted APIs, Actions, Pull Requests and Copilot

GitHub suffered a broad, worldwide platform disruption on August 17, 2026. The incident began at about 9:40 a.m. Eastern Time, or 13:40 UTC, and continued in some form until roughly 5:15 p.m. Eastern, or 21:15 UTC. It was not a total shutdown of every GitHub product, but it affected enough core services to interrupt repository access, code review, CI/CD automation, integrations and Copilot workflows.

What happened during the GitHub outage?

GitHub initially reported performance problems affecting some services. The incident then expanded across the website, API traffic and several developer-facing systems. The platform reported approximately a 20% error rate for web experiences and API traffic, while archive downloads and raw repository-content downloads experienced an approximate 50% error rate.

Those figures describe failed or unsuccessful requests, not the percentage of users who were unable to use GitHub. The available reporting characterizes the incident as global, but it does not provide a verified total number of affected users.

Which GitHub services were affected?

The outage reached multiple layers of the development workflow:

• Repositories and downloads: Archive downloads and raw files were especially unreliable, making it difficult to retrieve code through common web and automated access paths.
• API requests: Elevated API errors disrupted tools, scripts and integrations that depend on GitHub programmatically.
• Pull Requests and Issues: Degraded availability affected code review, issue tracking and other collaboration tasks.
• GitHub Actions: CI/CD workflows and automation were affected as Actions experienced degraded availability.
• Webhooks: Unreliable event delivery could delay or interrupt connected deployment, monitoring and automation systems.
• Git operations: Status reporting also recorded degraded Git performance, including problems affecting repository operations.
• Copilot: GitHub separately marked Copilot as a major outage or degraded service during the incident, affecting AI-assisted coding for some users.
• Enterprise identity and provisioning: Contemporary reports indicated impact to SAML and OIDC authentication, SCIM and Team Sync-related functions, although the available evidence does not provide a complete service-by-service timeline.

In practical terms, the incident was more than a website problem. A developer could encounter failures while opening or downloading repository content, reviewing a pull request, triggering or waiting for an Actions workflow, receiving a webhook event, authenticating through an enterprise identity setup or using Copilot.

Did every GitHub service go offline?

No. The evidence describes a partial, multi-service outage rather than a confirmed failure of every GitHub component. One contemporaneous status summary reported Git Operations, Packages, Pages and Codespaces as operational while other services were degraded.

That distinction matters: an operational status for one component did not guarantee that a user's broader workflow would function. For example, Codespaces or Pages could remain available while repository downloads, Actions, Pull Requests or webhooks were unreliable. The available reports also do not establish that any listed service remained fully unaffected throughout the entire incident.

What did GitHub say about the investigation and recovery?

GitHub said it was investigating elevated performance problems, then reported that it had identified a problematic component and taken corrective action. Updates during recovery indicated strong signs of improvement, although some errors remained elevated while engineers continued monitoring and applying mitigations.

GitHub's status page later marked the GitHub.com incident resolved. A later update said engineers were still addressing sporadic Copilot authentication failures in some applications, while Copilot usage through the GitHub CLI and GitHub App was reported as unaffected at that point.

Was the root cause known?

Not in the material available for this report. GitHub described the identification of a problematic component and the mitigations used to restore service, but the contemporaneous coverage did not include a technical root-cause analysis. The status page said a detailed analysis would be shared when available.

It is therefore too early to attribute the outage to a particular database failure, deployment, cloud-provider incident or authentication defect. The evidence supports describing the cause as not publicly established at the time of the incident reports.

How did this compare with recent GitHub and Microsoft outages?

The August 17 incident was broader than several recent GitHub incidents that were limited mainly to Copilot or to a single product area:

• On August 1, GPT-5.6 Luna in Copilot experienced intermittent availability problems tied to an upstream model provider; other Copilot models were not affected.
• On August 10, GitHub reported a Copilot disruption affecting access to some models.
• On August 13, GitHub reported errors involving the Fable 5 model in Copilot and attributed the issue to an upstream model provider.
• A separate August 6–7 GitHub Actions incident affected workflow availability, with failures and extended queueing during the event.

Unlike those narrower incidents, the August 17 outage spread across GitHub's web experience, APIs, collaboration features, automation, webhooks, repository access and Copilot.

Although GitHub is owned by Microsoft, the available evidence does not establish that this was a Microsoft 365 or Azure-wide outage. It is best described as a GitHub platform incident unless a later Microsoft or GitHub investigation demonstrates a shared underlying cause.