GitHub hit by widespread outage, breaking builds and logins for hours

Key points

• GitHub confirmed a widespread outage at 9:40 AM EDT on August 17, 2026, affecting the website, API, Actions, Pull Requests and more.
• Error rates hit roughly 20% across the web experience and API traffic, with repository archive and raw content downloads failing about 50% of the time.
• Login-related services including SAML, OIDC, SCIM and Team Sync were degraded, locking some users out.
• GitHub Copilot, the company's AI coding assistant, joined the outage list at 10:31 AM EDT.
• GitHub has not said what caused the incident and the investigation is ongoing.

GitHub, the Microsoft-owned site where millions of developers store and share software code, went sideways on Monday morning. For several hours, people trying to open a project page, log in, or run automated tests hit errors.

The company confirmed the trouble at 9:40 AM EDT on August 17, 2026, saying it was looking into performance problems across some services. Within the hour the list of broken things had grown considerably.

What actually broke?

Almost everything developers touch day to day. GitHub's own status page put the error rate at about 20% across the website and its API, which is the connection other tools use to talk to GitHub automatically.

Downloads were worse. Archive downloads and raw repository content, the plain files inside a project, were failing about half the time.

Logins were a big part of the pain. GitHub said SAML and OIDC authentication, along with SCIM and Team Sync, were all degraded. Those are the systems large companies use to sign staff into GitHub through their corporate accounts. When they wobble, whole engineering teams get locked out at once.

Users reported server errors when opening the site, and broken pages for commits, repositories and Pull Requests, the workspaces where code changes get reviewed before going live.

GitHub Actions, the service that automatically builds and tests code every time a developer saves a change, was running slowly. That means release pipelines at countless companies stalled. At 10:31 AM EDT the company added GitHub Copilot, its AI coding assistant that suggests code as you type, to the list of degraded services.

Git Operations, Packages, Pages and Codespaces were still listed as working, though several other core pieces were not.

Was this a cyberattack?

There is no evidence of one. GitHub has not disclosed a cause, and outages of this shape, spread across many internal services at once, most often trace back to a configuration change, a database problem, or a shared internal dependency failing. That is speculation, not attribution.

As first reported by BleepingComputer, the company's investigation was still ongoing at the time of publication. Until GitHub posts a post-incident review, treat any cause floating around social media as unconfirmed.

What should developers and companies do?

Wait it out, and check the status page rather than retrying blindly. Hammering a struggling service with retries usually makes recovery slower.

If your continuous integration jobs, the automated checks that run on every code change, failed during the window, re-run them once GitHub reports green. If your staff could not sign in through single sign-on, that was the SAML and OIDC issue, not their password.

The wider lesson is a familiar one. A very large amount of modern software depends on a small number of platforms staying up. When GitHub sneezes, release trains at banks, hospitals and startups all pause together.

This is a developing story and will be updated as GitHub shares more.