If you had read the "lengthy AI essay", you would have noticed that this was the exact thing I criticized in another AI MCP approach. And IMHO it has a lot to do with the topic, namely that we all think that there are many updates that cause a lot of manual work.I understand that this topic is important to you. However, it is only very indirectly related to the subject of this thread. The discussion here is about whether there are too many kernel updates and how these could be reduced (presumably mainly because people don't want to reboot, and partly because they're worried about regressions).
Also, and this is more of a side note, independently of whether the improvements you're asking for are actually needed (I only skimmed your two lengthy AI-generated essays), if you're concerned about operational safety, I'd advise caution when letting non-deterministic LLMs loose on your systems.
I also don't think it's a good idea to completely automate updates on critical infrastructure systems like hypervisors. Simply restricting access to this feature to selected admins, allowing them to perform updates manually instead of potentially letting AI agents mess up your hypervisors, would also somewhat mitigate the risks associated with having a root shell in the UI.![]()
My take on this is:
- Having security updates is better than having none, so do not complain about "too may updates".
- The reason for the flood is AI-detected exploits and this is not gonna end soon.
- If you have a big exposure, the mitigate downtimes by HA and automatic VM migration.
- If you want up lessen the burden of upgrading lots of PVE and PBS machines, you might want to use an automated approach.
- That approach might ironically be using AI yourself OR any API-based approach. That being said, I would not trust the AI by exposing the raw API, but rather encapsulated isolated work flows via an MCP server or even a script using the API to upgrade and potentially reboot the server.
- I found that to be infeasible REGARDLESS of if you want to use AI. It is the API itself that is lacking the feature.
- If you want to circumvent that, you are actually increasing the attack surface, because you have to use a full shell to do the updates.
Last edited: