Nvidia L40S on Windows Server 2025 and WSL2: Failed to initialize NVML: GPU access blocked by the operating system

Nobody has claimed this yet.

### Windows Version

Microsoft Windows [Version 10.0.26100.32370]

### WSL Version

2.6.3.0

### Are you using WSL 1 or WSL 2?

- [x] WSL 2
- [ ] WSL 1

### Kernel Version

Linux version 6.6.87.2-microsoft-standard-WSL2 (root@439a258ad544) (gcc (GCC) 11.2.0, GNU ld (GNU Binutils) 2.37) #1 SMP PREEMPT_DYNAMIC Thu Jun 5 18:30:46 UTC 2025

### Distro Version

Ubuntu 24.04

### Other Software

_No response_

### Repro Steps

Windows Server 2025, installed Nvidia drivers (latest version): 32.0.15.9159 in Data Center mode: WSL doesn't see Nvidia L40S card.
Under pure Windows ollama.exe works, in powershell nvidia-smi reports working of card as it should but when I try to use nvidia-smi it returns:
```
Failed to initialize NVML: GPU access blocked by the operating system
Failed to properly shut down NVML: GPU access blocked by the operating system
```
When trying to start any docker image containing AI workload it reports: CUDA not available

### Expected Behavior

Expected:
- regular nvidia-smi output
- CUDA found by eg. WebUI

### Actual Behavior

```
$ nvidia-smi
Failed to initialize NVML: GPU access blocked by the operating system
Failed to properly shut down NVML: GPU access blocked by the operating system
```

### Diagnostic Logs

[WslLogs-2026-03-03_07-48-34.zip](https://github.com/user-attachments/files/25707880/WslLogs-2026-03-03_07-48-34.zip)

Start by reviewing the attached WslLogs-2026-03-03_07-48-34.zip and the reported WSL 2, kernel, Ubuntu, and NVIDIA driver versions. Reproduce the nvidia-smi failure and check whether the same GPU access problem affects the mentioned Docker AI workload. Done means nvidia-smi returns regular output and CUDA is available to the workload.

Written by the indexing model from the issue text.