Home Creators Posts Import Register Favorites Logout

Downloads

Content

Hey everyone!

Today, we have:

00:05 PostmarketOS changes name

https://nura.eco/blog/2026/09/27/nura-rename/

02:40 Firefox 157 adds the new Nova theme

https://linuxiac.com/firefox-157-nova-released-with-its-biggest-visual-refresh-in-years/

06:24 Nvidia announces new layer to confine AI

https://nvidianews.nvidia.com/news/open-agent-safety-platform

08:40 Dev proposes that GNOME accepts AI contributions

https://arunraghavan.net/2026/09/on-llms-and-free-software/

have a great day!

Files

Comments

UsernamesAreHard

Imagine giving the keys to your car to a toddler and then installing a camera to monitor that the toddler doesn't get into your car. Is installing a camera the right solution to prevent the toddler from damaging your car? Maybe not giving the car's keys to the toddle in the first place would be a smarter solution, no? Yet here we are, giving Ai write access to the world and putting in place bollocks "sandbox" "solutions" to watch over said AI instead of behaving responsibly and not letting Ai have opportunities to act in our world. We live in an idiocracy.

Nick - The Linux Experiment

I can understand trying to "contain" an AI inside a sandbox, I think it should even be the very default on all of these tools, but yeah, it feels like rushing towards deploying these things as fast as possible, and figuring out the security later, which is a weird way to tackle the issue. At that point, I think the only way these things should even remotely be usable should be to run a local model inside of a container that has 0 access to the actual OS you're using.

UsernamesAreHard

While I agree that by default ai models should be contained and sandboxed somehow, I think that it is laughable to think that sandboxing agents is a sufficient security measure. What scares me is not how the ai is sandboxed but what the ai agent is asked to do and what tools the agent is given. If you ask an ai agent specialized in pentesting to complete given cyber security tasks in an isolated sandboxed environment, is it realistic to expect the agent to not break lose? This is what happened with OpenAi agents when Hugging Face got hacked this summer. If you're sandboxing the agent so that it can't touch the rest of the system it's running from, what does the sandbox protect from if the agent can wreak havoc on a remote system? Protecting the system it's running from is probably the least of the concerns if what the agent is controlling is more critical. Imagine letting an Ai agent control the temperature of a nuclear reactor on its own. What good does sandboxing said agent do when the agent could decide to just let the reactor melt, if it's given that unsupervised power? That's why I think that it is ridiculous that this "sandbox" by Nvidia is not more than a well wishing self pat on the back.

Lerothas

Theming is one of the main reasons I avoid Firefox and Thunderbird as much as possible - and of course GNOME apps. If I'm okay with a weird mix of design languages I would use windows or macOS. But Linux has more to offer. So I'm really sad Firefox, Thunderbird and all Chromium based browsers are ignoring that at all. 😭

Clint Eschberger

The most important thing Firefox did was un-deprecate compact mode! 😎