Writer: Jordan Sprogis
Editor: Lillian Castro
Reviewer: Cristian Lopez
Key Takeaways
GitHub finished investigating an issue reportedly involving Copilot’s policies at around 10 p.m. UTC on Feb. 9. The issue had been ongoing since midday and affected the AI coding assistant wherever it was integrated, including VS Code and GitHub itself.
According to the official status page, GitHub said some of Copilot’s policy changes failed to apply as they normally would. Everything is operating normally now, and GitHub hasn’t said what caused the issue.
Just as a refresher, a Copilot policy update is when end users of the API can adjust the rules that control how Copilot is used in the end user’s system, such as which teams have access or the repositories Copilot can see.
Users floated the idea on developer forums that new Copilot models showed up before enterprise policies had finished applying.
“New models are published to the drop down and it breaks copilot. Looks like AI was allowed to push this update without any review,” wrote one Redditor on r/GitHubCopilot.
GitHub’s status update noted similar symptoms, saying newly enabled models didn’t show up correctly for some users. Oops.
It’s happened before on GitHub — some past incidents affected Actions and Codespaces and, while it didn’t take them completely offline, it did interrupt some essential processes. Last week, some devs also reported workflow interruptions on Actions.
“Haven’t been able to run any actions for the last 30 mins… also had two in progress that were cancelled,” wrote one user on r/GitHub.
For something so small, it hits at an incredible scale: Nearly 85% of developers report using AI tools at work, and among those users, around two-thirds say they actively use Copilot. Copilot is also used by 90% of the Fortune 100 companies, with one analysis suggesting it has between 4 million to 6 million daily users.
It’ll happen again. It’s just par for the course. This time, a decent chunk of the world’s developers who may have had a bit of a slow day were having to say: “Sorry, couldn’t finish that website for you. My vibe coder was down.”
