Internet Status — August 17, 2026: GitHub Outage Recovering

GitHub spent most of Sunday working through a critical incident. Starting at 13:40 UTC, error rates hit roughly 20% across the GitHub.com website and API, with archive and raw-file downloads failing at closer to 50%. Pull Requests, Issues, Actions, Webhooks, Git Operations, Pages, Copilot, and SAML/OIDC sign-in were all degraded. GitHub identified the problematic component and applied corrective actions, and most services recovered between 17:00 and 19:00 UTC — but the incident was still open as of 20:20 UTC, with sporadic authentication failures lingering for Copilot in some applications (githubstatus.com).

AWS is in the final stretch of a multi-day repair in Frankfurt. A facility infrastructure failure at the Equinix FR5 colocation site has impaired AWS Direct Connect connectivity in eu-central-1 since August 15. As of 18:00 UTC today AWS reports broad recovery, with the majority of affected hardware restored and only a small number of customers still impacted; customers with redundant Direct Connect locations were never affected (health.aws.amazon.com).

Elsewhere, a cluster of smaller issues: Vercel had a partial disruption of its dashboard, observability, analytics, firewall, and login from 14:30 UTC — services recovered by about 16:45 UTC and missing observability data is being backfilled (vercel-status.com). Cloudflare's Lists service is degraded (changes to URL Lists delayed; existing lists and the edge network unaffected) per cloudflarestatus.com. Twilio is seeing elevated Verify Silent Network Auth failures for AT&T users in the US (status.twilio.com). Slack is still working a lingering issue from August 13 where users with custom data-retention policies may be unable to load older messages (slack-status.com).

Core infrastructure is otherwise healthy: AWS (outside Frankfurt), Google Cloud, Azure, Azure DevOps, Akamai, Stripe, OpenAI, Discord, Atlassian, Zoom, Heroku, and DigitalOcean all report normal operations. No BGP, DNS, or undersea-cable incidents detected. Fastly's status page was unreachable from our monitoring vantage, but no third-party signals suggest any Fastly issues.