Gemini desktop leak shows Task mode and app controls
Published screenshots show a task selector and computer use settings. The September 24 report leaves actual operation and public availability unverified.

Gemini’s desktop app is testing a task interface with more explicit computer use controls, according to TestingCatalog’s September 24 report. The outlet says trusted testers were beginning to receive access, but states that it did not have access to the mode itself.
Its screenshots show a Chat and Task selector, a voice control and settings for apps that are always or never permitted. The settings explicitly refer to a Mac. ByteForward reviewed the published images, but has not authenticated the build or tested those controls.
Controls are visible while operation remains unverified
A permission screen is useful evidence of the experience being designed. It does not demonstrate that restrictions are enforced, which actions require approval, or whether a user can safely interrupt unfinished work. Those are separate questions that would need documentation and testing.
The voice control also cannot establish that a user can direct desktop actions through a live conversation. The report suggests that possibility, but supplies no verified demonstration of that complete workflow.
What Google has actually announced
Google released its Windows app on September 10 for Windows 10 and 11. Its announcement describes opening Gemini with Alt and Space, using existing Gemini services and creating images or videos. It also says more native desktop capabilities will arrive over time. That public Windows release should not be treated as confirmation that the pictured Mac controls are generally available.
Google separately introduced Gemini 3.8 Live on September 15. The company says the model can execute tools and API calls in the background while continuing a conversation. That is relevant technical context, but the model announcement does not verify how the leaked desktop interface uses it.
For anyone evaluating a desktop assistant, the material questions are practical. Which apps can it reach, what approval is required, and what happens when a task fails midway? The reported settings make those questions worth asking. They do not yet provide the answers or establish a public launch date.
Original editorial illustration created for ByteForward with AI. It is conceptual artwork rather than a product screenshot.



