Windows Sandbox is supposed to be a clean, disposable environment that “just works.” But every so often, it bites back. In the past few days I’ve hit a recurring issue where Sandbox loses network access, stalls, or refuses to start after a failed session. The symptoms vary, but the root cause keeps pointing to the same culprit: corrupted HNS state. This post walks through what happens, why it happens, and how to fix it without drama. Alas, it means I’ve been bitten by the Sandbox network access bug widely reported these days (see Windows-Sandbox and Windows-Containers issues in GitHub; 131 and 128 for the former, and 631 for the latter).
Sandbox Network Access Bug — TLDR Summary
When Windows Sandbox suddenly loses network access or fails to relaunch, the underlying HNS (Host Networking Service) stack gets stuck in a half‑deleted state. Resetting the HNS networks and rebooting sometimes restores normal behavior. But it keeps coming back with each use, and the fix doesn’t always work.
How the Bug Shows Up
The first sign is usually a sudden loss of connectivity inside Sandbox. A browser tab stops loading, a site hangs, or a download stalls. Closing Sandbox seems harmless enough, but reopening it often triggers a second symptom: the app refuses to start. Sometimes it crashes, after which Windows throws error 0x80370106. Other times it simply flashes “connection lost” and drops you back to the desktop.
On my Flo6 machine, the failure pattern became predictable. Sandbox would run normally for a while, then lose network access. After closing it, the next launch attempt failed. Meanwhile, my P3 Ultra Gen 2 workstation ran Sandbox without complaint. That contrast made the diagnosis easier: the problem wasn’t the workload, the site, or the browser. It was the host’s virtualization networking stack.
Why HNS Gets Stuck
Windows Sandbox relies on a lightweight virtual switch managed by HNS. When Sandbox closes cleanly, HNS tears down the temporary network and resets its internal tables. But when Sandbox crashes or loses connectivity mid‑session, the teardown doesn’t complete. HNS keeps references to a dead network, and the next launch attempt hits a corrupted state.
Stopping HNS with net stop hns rarely works in this scenario. The service stays in a running state even while trying to unwind stale entries. The fix requires removing the broken networks directly.
The Reliable Fix
The repair workflow is simple and safe:
1. Remove the stale HNS networks using PowerShell:
Get-HNSNetwork | Remove-HNSNetwork
2. Reboot the system. This step matters because HNS rebuilds its network tables only after a restart.
3. Launch Windows Sandbox again. In most cases, network access returns immediately and the environment behaves normally.
This reset does not affect standard networking. Windows automatically recreates the default switch, NAT configuration, and Sandbox virtual network during startup.
Why This Keeps Happening
Sandbox is lightweight by design, but its networking stack depends on the same components used by Hyper‑V, WSL, and container workloads. Any hiccup in those layers can leave HNS in a confused state. Heavy browser workloads, WASM execution, or aggressive site behavior can push Sandbox into a crash that leaves the network half‑torn‑down.
Because the bug lives in the host’s virtualization plumbing, toggling Sandbox on and off in optional features doesn’t fix it. Only a full HNS reset clears the corrupted entries.
Final Thoughts
Windows Sandbox remains one of my favorite tools for quick testing, but the network access bug is an annoyance. The good news is that the fix is consistent and non‑destructive. If Sandbox bites you with a sudden loss of connectivity or a refusal to start, resetting HNS is the fastest way to get back to work.
Good thing it works just fine on the ThinkStation P3 Ultra Gen 2. When I hit issues with Sandbox on Flo6, I just remote into the P3 and keep going. Here in Windows-World, it’s always good to have an alternative workaround.

