Microsoft Copilot shows up even when unwanted

Microsoft customers are claiming the Windows giant's Copilot AI service sometimes ignores commands to disable the thing, and thus turns itself back on like a zombie risen from the dead.

A bug report to the corporation's Visual Studio Code, aka VS Code, Copilot repository from crypto developer rektbuildr claims Github Copilot enabled itself within various VS Code workspaces on its own.

Rektbuildr wrote, "I enable Copilot for specific windows, because not all my repos are public. Some belong to clients I work for and who did not consent to me sharing the code with third parties.

"Today Copilot enabled itself for all my open VS Code windows without my consent. I have agent mode enabled, so you now may or may not have a copy of all the files containing keys, yaml secrets, certificates and so on. That's not OK."

Microsoft did not immediately respond to a request for comment, though the biz appears to have assigned a developer to look into the issue.

The bug reporter also pointed to a post on Reddit describing how Windows Copilot had re-enabled itself on a PC after being disabled through a Group Policy Object setting.

In the ensuing discussion, an individual posting under the name kyote42 suggests this may be the result of a change in the way Microsoft implements Copilot on Windows 11.

"The GPO setting that disabled icon isn't valid anymore for the new app version of Copilot," kyote42 wrote. "Depending on your flavor of Windows 11, there are some steps outlined in the article, 'Remove or prevent installation of the Copilot app.'"

Uninstalling Windows Copilot now requires using PowerShell and then preventing reinstallation via AppLocker, according to Microsoft documentation.

Avoiding AI has become something of a challenge with other vendors, too. With the arrival of iOS 18.3.2 in March, Apple customers found that the update re-enabled Apple Intelligence, the iBiz's AI suite, for those who had previously tried to snuff it out.

• LLMs can't stop making up software dependencies and sabotaging everything
• AI spending spree continues as Microsoft commits $80B for 2025
• Despite Wall Street jitters, AI hopefuls keep spending billions on AI infrastructure
• DeepSeek means companies need to consider AI investment more carefully

Software developer Joachim Kurz claims that Apple's Feedback Assistant – used to pass bug reports to the iGiant – now includes a submission dialogue that states any submitted information can be used for AI training. (The Register was unable to reproduce this behavior using Feedback Assistant on macOS 15.4, but the dialogue may be present in more recent macOS releases.)

Google now forces AI Overviews on search users, whether they want it or not.

Meta AI, the chatbot service integrated with Facebook, Instagram, and WhatsApp can't be turned off either, though there are some partially effective ways to limit it. Meta recently said it will be harvesting public social media posts of Europeans for AI training unless they opt-out.

Mozilla has taken a more nuanced approach. The public benefit biz has been shipping an AI Chatbot sidebar since Firefox 133 and presently requires the user to activate the sidebar and configure it with an AI model. Nonetheless, a pull request to remove the feature from the Zen browser, which is a fork of Firefox, suggests even that a kinder, gentler approach is unwelcome in some quarters.

DuckDuckGo allows users to choose to avoid AI or not. It offers the subdomain noai.duckduckgo.com to load its search engine without an AI chatbot icon and its standard domain duckduckgo.com to load a page with AI.

But overall, the creeping AI encroachment is getting harder and harder to avoid. It might have something to do with the billions these super-corporations have sunk into the technology. ®