Amutable Foundations: A Technical Overview

[Posted November 7, 2026 by corbet]

Lennart Poettering, Michael Vogt, and Joaquim Rocha have put together a detailed overview of what the Amutable Operating System (AOS) and associated infrastructure will look like.
The image-based OS approach enables us to enforce some very strict but powerful security properties, which elevate the security posture of the system substantially above traditional Linux systems. Specifically, let's highlight seven major essential security properties of AOS here:
  1. No Shell + No SSH + No Language Interpreters. By default AOS does not install a UNIX shell, nor makes remote access via SSH available on the nodes. Moreover, we do not install any of the usual language interpreters (Python, Perl, …) by default. For most nodes a shell or interpreted language support is not required and it removes a whole class of attacks.
  2. Signed Images Only. On AOS only images that come with Verity data signed by a signing key listed in the Verity keyring may be activated.
  3. [...]


to post comments

Moving the game forwards

Posted Nov 7, 2026 9:43 UTC (Sat) by alspnost (guest, #2763) [Link] (4 responses)

I like this concept, broadly speaking. Is there a reason why I shouldn't? And is that reason "because it's Agent P" or something else?

Moving the game forwards

Posted Nov 7, 2026 13:53 UTC (Sat) by hailfinger (subscriber, #76962) [Link] (3 responses)

Moving the game forwards

Posted Nov 7, 2026 16:18 UTC (Sat) by elru (subscriber, #170791) [Link] (2 responses)

Moving the game forwards

Posted Nov 7, 2026 22:19 UTC (Sat) by hailfinger (subscriber, #76962) [Link] (1 responses)

Moving the game forwards

Posted Nov 7, 2026 23:55 UTC (Sat) by elru (subscriber, #170791) [Link]

Factory reset

Posted Nov 7, 2026 12:09 UTC (Sat) by alx.manpages (subscriber, #145117) [Link] (9 responses)

Factory reset

Posted Nov 7, 2026 12:40 UTC (Sat) by NekkoDroid (guest, #176638) [Link] (4 responses)

I am unsure if AOS does it like this, but what Lennart has always been working towards is that the UEFI verifies the kernel (UKI) using the public key, which then verifies the /usr/ partition (which is read-only), which then loads the rest. "the rest" here would be the factory reset part, which is the only really modifiable part (also EFI variables get reset).

So say the nation state enrolled some kinda public key into the mutable part of the system (/var/ or possibly /etc/ but it sounds like it is also read-only) so that the system now also loads some of their system extensions or so, a factory reset would basically wipe those keys (and the extensions). /usr/ can't be enrolled, since that is pinned by the verity root hash and changes in the data would cause errors during reading by the kernel and the UKI is "pinned" by you (or I guess in the case of Amutable them) signing new UKIs. This is assuming you can prevent the nation state from enrolling their keys in your UEFI. But even if they did enroll their own keys in the UEFI the PCR values would be different, resulting in the mutable LUKS partitions not being automatically unlocked, so your mutable data is still inaccessible.

So as long as the cryptography implementations are sound and not bypassed by some 0-days it should be sound for the most part. Unless they are able to get below the bootloader it should be fine (but at that point they don't even need your PIN).

Do tell me if I maybe missed some attack vector which could possibly still exploit something.

Factory reset

Posted Nov 7, 2026 13:57 UTC (Sat) by hallyn (subscriber, #22558) [Link] (2 responses)

Factory reset

Posted Nov 7, 2026 15:53 UTC (Sat) by NekkoDroid (guest, #176638) [Link] (1 responses)

Do note that pretty much all desktops also have co-processors (Intel ME and AMD PSP).

Imma cop-out and say that is part of "getting below the bootloader" :) If you can't trust the hardware you are running on it is very difficult to trust anything running on top of that. If the hardware vendor "sells you out" you are fucking anyway I would say.

Other than usually signed firmware, is there any way to even persistently choose which code is run on the coprocessors?

Factory reset

Posted Nov 7, 2026 16:08 UTC (Sat) by alx.manpages (subscriber, #145117) [Link]

Factory reset

Posted Nov 7, 2026 16:46 UTC (Sat) by ballombe (subscriber, #9523) [Link]

They replace the UEFI.

Factory reset

Posted Nov 7, 2026 16:13 UTC (Sat) by hailfinger (subscriber, #76962) [Link] (2 responses)

Factory reset

Posted Nov 7, 2026 16:20 UTC (Sat) by alx.manpages (subscriber, #145117) [Link] (1 responses)

Factory reset

Posted Nov 7, 2026 23:17 UTC (Sat) by jepsis (subscriber, #130218) [Link]

What is the manufacturer of your phone? And also the model if you can tell it?

Factory reset

Posted Nov 7, 2026 18:41 UTC (Sat) by NightMonkey (subscriber, #23051) [Link]

No Browser?

Posted Nov 7, 2026 14:08 UTC (Sat) by welinder (guest, #4699) [Link] (4 responses)

No Browser?

Posted Nov 7, 2026 15:27 UTC (Sat) by tux3 (subscriber, #101245) [Link] (1 responses)

It doesn't look like this is aimed at desktop users. I have some k8s nodes that my cloud provider installed Ubuntu on, so I get to enjoy broad compatibility with all the latest CVEs for up to a week. In an ideal world, those nodes don't need a browser, SSH, or a shell at all.

But I'm still very happy on my Debian at home. And actually, most Debian packages turn out to not meaningfully depend on install order, most don't have non-trivial scripts, or only depend on simple triggers. I have a project where I turn Debian packages into erofs layers as a performance hack, and the non-reproducibility of packages is a bit overstated. It's limited to a few things like shared caches updated by triggers which are not frozen by design, but apart from that most of the bits on disk can be made reproducible without having to throw the whole packaging system away, if you so desire.

No Browser?

Posted Nov 7, 2026 16:59 UTC (Sat) by davidstrauss (subscriber, #85867) [Link]

Correct. We are not aiming for the desktop.

No Browser?

Posted Nov 7, 2026 18:53 UTC (Sat) by AdmiralNemo (subscriber, #113495) [Link] (1 responses)

Wouldn't it also preclude Polkit, since it needs a JavaScript runtime for configuration?

No Browser?

Posted Nov 7, 2026 19:09 UTC (Sat) by bluca (subscriber, #118303) [Link]

No, since the switch to duktape as the engine there's no more JIT

Is there a comparison with alternatives like Talos or Bottlerocket?

Posted Nov 7, 2026 19:43 UTC (Sat) by ejr (subscriber, #51652) [Link]

There are a bunch of design decisions that could be highlighted with a clear comparison. I suppose also things like k3os. I poked around the website and didn't see anything obvious.

内容来自 LXer Linux News,版权归原作者所有。

阅读原文 ↗

把喜欢的内容,留在你的新标签页。

打开我的订阅