GitHub Copilot Models Down
Most of the internet is running normally this morning, but GitHub is carrying an active, high-impact incident.
**GitHub Copilot AI Model Providers — ongoing.** GitHub opened an incident at 10:04 UTC reporting degraded availability for Copilot AI Model Providers, and marked the component as a major outage. The incident is still in the investigating stage with no mitigation announced yet, so Copilot code completions and chat may fail or time out for some users. GitHub rates the impact as critical. Live updates: https://www.githubstatus.com
**GitHub Billing — ongoing, minor.** A separate billing incident opened at 23:37 UTC on August 26. Billing budget pages failed to load and Copilot CLI sessions failed to start or continue. GitHub applied a mitigation that restored Copilot usage and is now monitoring the remaining billing page disruption.
GitHub also resolved two shorter incidents in the last 48 hours: a degradation affecting Actions and Pull Requests earlier today, and an Actions incident on August 26 that GitHub traced to database saturation.
**Everything else is quiet.** Google Cloud reports no broad severe incidents. Microsoft Azure reports no active events. Cloudflare has no open incidents — only scheduled datacenter maintenance in Ashburn, Philadelphia and Hamburg — after resolving a roughly 90-minute Durable Objects degradation on August 26 and a brief Realtimekit API failure earlier today. Akamai, OpenAI, Anthropic, Discord, Atlassian, Zoom, Vercel, Heroku, DigitalOcean and Stripe are all operational, as are Cursor, Groq, Replicate, ElevenLabs, Cohere and Stability AI.
Two smaller items worth noting. Slack has a long-running minor incident, open since August 13, affecting access to historical messages in workspaces with custom data retention policies. Twilio is reporting several carrier-level regional issues — SMS delivery delays to ICE subscribers in Costa Rica, MMS delays to C-Spire in the US, delayed delivery receipts in El Salvador, and voice call problems in Sweden and Brazil that are now recovering. These are localized to specific carriers rather than platform-wide.
As standing background, AWS still lists its two Middle East regions — UAE (me-central-1) and Bahrain (me-south-1) — as unable to reliably support customer workloads following damage sustained earlier this year. That is a long-term condition, not a new development, and AWS has advised customers to recover resources in other regions. No new AWS incidents are open elsewhere.
We found no evidence of BGP routing incidents, DNS failures, undersea cable faults, or country-level connectivity loss today. Crowd-sourced outage trackers show only routine, localized ISP and consumer-service noise.
Overall: incidents detected, but no widespread internet disruption. The main thing to watch is the GitHub Copilot model provider outage, which is still unresolved as of publication.