Claude Desktop Bug Reportedly Leaves a 1.8GB Hyper-V VM Running on Windows

Anthropic has been making headlines recently following its reported 965 billion valuation and the launch of Claude Fable 5 and Mythos 5. However, a much smaller issue has started attracting attention among Windows users of Claude Desktop.

A GitHub issue originally opened in February has resurfaced after users reported that Claude Desktop on Windows may continue running a Hyper-V virtual machine even after agent-based features are no longer being used.

According to reports from users, the problem appears after Claude Cowork, also known as agent mode, is launched at least once. After that, a Hyper-V virtual machine reportedly starts every time Claude Code is opened through Claude Desktop, even when users only want to use standard chat features.

Users report a persistent Vmmem process and increased memory usage

Affected users say they can see a Vmmem process in Windows Task Manager after launching Claude Code.

While the process reportedly shows little or no CPU activity, it continues to reserve approximately 1.8GB of system memory. Some users argue that the virtual machine should only start when agent mode is actively required and should shut down automatically once those tasks are complete.

The issue appears limited to Windows desktop installations. Reports suggest that users running the Claude CLI directly, as well as users on other platforms, do not experience the same behavior.

Several comments on the GitHub thread also point to concerns about session cleanup. Users believe Claude should remove temporary resources associated with agent mode after a session ends rather than repeatedly launching VM infrastructure for normal chat interactions.

No official fix or workaround yet

At the moment, the exact cause of the behavior remains unclear. It is also unknown when the bug first appeared, although the GitHub report has reportedly remained open for several months.

Users have requested several improvements, including launching the Hyper-V environment only when agent mode is explicitly enabled, properly cleaning up session files after use, allowing Claude to work gracefully even when VM-based infrastructure is unavailable, and ensuring that the issue does not affect normal chat performance.

So far, there is no confirmed workaround for the issue. Anthropic has also not publicly indicated whether it plans to investigate or prioritize a fix.

Recently, Anthropic also drew attention after suggesting that future AI development could potentially require global coordination or even temporary pauses under certain circumstances.

For now, Windows users affected by the issue may need to wait for further clarification from Anthropic regarding the cause of the persistent Hyper-V virtual machine behavior.

Via Neowin