I'm confused why VM + systemd-nspawn? From my understaing WSL 2 runs a single VM + something like systemd-nspawn per "linux installation", but it runs a VM because it needs linux kernel. Why not just do systemd-nspawn if you alread on linux?
Way better isolation, is my guess. Plus, you can use a different kernel this way.
I used to poo-poo when people said that containers aren't a _real_ security boundary, at least for personal stuff, and not a multi-tenant server. But I bet even mid-tier LLMs can break out of LXC/Docker/nspawn at this point.
That has to be the host kernel not the guest kernel. Old Nvidia systems without supported kernel drivers often lack IOMMU to forward PCIe memory, too. I wouldn't recommend using an outdated host kernel with 2026 AI-powered vulnerability scanners.
Windows (10 LTSC or 11 with dTPM) actually works better for old Nvidia systems with WSL. You even get CUDA libraries within WSL and security updates for the next 5 years.
CVEs for runc are much more frequent than CVEs for KVM. The attack surface area is bigger, and containers were never intended as a security boundary, but rather as a resource management tool.
it is quite likely that you can break out even with no linux containers related CVEs. --isolate does seem to fix that somehow but is explicit opt. in and "more painful to use" ... (which creates a UX challenge unlikely to end well from a security POV).
I was just testing this and it's not clear that it works out of the box. `krun` shows a different kernel than with `crun` but it doesn't reflect the dropped capabilities in the same way. I'm probably holding it wrong but I'm not sure what to look for at the moment.
"During a test conducted by Trail of Bits researcher Artem Dinaburg, a preview version of GPT 5.6-Cyber was tasked with breaking out of a Debian 12 virtual machine. Initially, the agent exploited a known Linux kernel vulnerability, CVE-2026-53359, by developing its own exploit. After the host was updated, the agent found another pathway through libslirp, chaining a known vulnerability (CVE-2026-9539) with a previously unassigned bug to gain arbitrary host memory access. Even after QEMU and libslirp were updated, the agent analyzed system components and constructed a new escape chain using three zero-day vulnerabilities and one KVM flaw that had not yet reached the distribution kernel.
These findings suggest that general-purpose VMs may not be adequate security boundaries for highly capable AI agents, especially in older systems with delayed security updates. Trail of Bits recommends using specialized isolation systems like Firecracker, restricting VM access, and implementing rapid patching to mitigate these risks."
Can confirm. My abliterated models kept breaking out of qemu VMs due to BIOS implementation quirks in qemu.
I'm using firecracker now with a very defensive systemd-as-separate-non-admin-user seccomp sandbox on top, which seems to hold them off long enough for me to see an agent going rogue and intervening.
Currently I still have hopes that eBPF sandboxing will help, but just a couple days ago my agent discovered a use after free bug in the ebpf kernel-side verifier... so there's that.
I was surprised I hadn't heard of this as a separate tool from systemd-nspawn. Then I realized why... the GitHub repo shows version 0.6 released in 2022, then version 1.0.0 last week, followed by a flurry of releases up to 1.8.0 yesterday.
So basically, it's been all Claude'd up extremely recently.
That said, the landing page, docs, and git README are much higher quality than I normally see out of LLM-generated projects, so at least the author knows how to reign in the needless verbosity and write for a technical audience. So I will give him credit for that at least.
nspawn maintainer here. Yes, it has been done with the help of AI (the README has a clear AI usage disclosure), but I have been writing Rust code since a very long time ago, before agentic coding was a thing (you can check my GH profile). Every change is reviewed by me, and the architecture/implementation is dictated by me as well.
Hi, nspawn maintainer here, that's not the case. I have been working with code/software architecture/design since a very long time before agentic coding was a thing. nspawn uses AI to help write code (the README has a clear AI usage disclosure) but every change is reviewed by me.
The architecture decisions, etc are dictated by me as well :)
If you have any suggestions, etc, they are welcome.
I'd trust OP to make good architectural decisions/provide guidance even if AI mostly wrote the docs and UI. Looking into their GitHub, seems they are an engineering manager at Microsoft and contribute semi regularly to uBlue and Project Bluefin. Should be better quality than some random vibeslopper.
