Anthropic has recently dominated headlines, not only for its impressive revenue and valuation updates but also for unveiling its cutting‑edge Fable model. While the company’s flagship chatbot, Claude, continues to attract millions of users, a growing number are reporting frustrating memory glitches when they engage with the platform.
On the Claude Code GitHub page, a February‑posted issue has resurfaced with renewed urgency. The problem stems from Claude Desktop on Windows, which launches a 1.8 GB Hyper‑V virtual machine whenever users activate Claude Cowork or agent mode—even if they only intend to use the app in simple chat mode. This unintended VM starts up on every launch of Claude Code, regardless of the chosen interaction style.
Many users have upvoted the bug, confirming that the issue plagues their machines as well. Interestingly, the glitch appears to affect only Windows desktop users; those who rely on the CLI or other platforms seem untouched. Once triggered, the Vmmem process appears in Task Manager, showing a CPU usage of 0 % while hogging an astonishing ~1.8 GB of RAM.
Claude’s community is demanding that the Vmmem process only activates when users explicitly launch agent mode or Cowork, and that session files be efficiently purged afterward. They also call for a graceful fallback when virtual‑machine infrastructure is unavailable, all while maintaining the high‑performance chat experience that Claude is known for.
The origin and underlying cause of the recent glitch remain uncertain, yet the community has once again turned to GitHub and Hacker News to discuss the problem. Additional technical insights and log entries are also available on the GitHub repository.
It is still unclear whether Anthropic will investigate the issue, particularly since it has been reported for several months. Nevertheless, the bug is inflicting significant frustration on users, with many reporting that it has forced them to uninstall Claude Code on desktop due to the absence of a definitive workaround.
News Source: Neowin
Comments
Be the first to comment.