Recently GamingOnLinux highlighted Canonical's plans for adding AI features into Ubuntu Linux, and naturally this has caused plenty of concern.
People are right to be concerned, as we've all seen the horror stories about AI going rogue and doing all sorts of stupid things. From deleting entire databases, to security issues - there's a lot to think about. Providing some clarifications on what the plans are going forward, Jon Seager, the VP Engineering for Canonical, replied to the original post noting some important bullet points about it:
- On the idea of a kill switch: while I said that we won’t add a “global kill switch”, all of these capabilities will be delivered as Snaps to the OS, layered on top of the existing Ubuntu stack. That means there will always be the option of removing those Snaps - which I suppose acts as a sort of kill switch for the features we’re planning on shipping.
- Opt-in vs Opt-out: my plan is to introduce AI-backed features as a “preview” on a strictly opt-in basis in 26.10. In subsequent releases, my plan is to have a step in the initial setup wizard that allows the user to choose whether or not they’d like the AI-native features enabled. Because of the size of most LLMs, we simply couldn’t ship them in the installer anyway, so opting out at first run is simple: they just won’t be there.
- On cloud providers: there appears to be some concern about “sending logs to the cloud” and such. To be clear, this will not be part of our plans. Default configurations of these tools will always be to use local inference against local models. In order to use cloud-based inference, you would need to explicitly configure that, and provide an API token or other credential
At least this seems like a sane plan, if there are going to be AI features available. Not there by default, fully opt-in and easy to remove. Making it easy for both sides to deal with: those who want it, and those who do not.
What about Canonical shipping code that was created or co-authored by AI? That will be a thing in Ubuntu. Seager mentioned "in reality we will be doing this", backing it up by mentioning how "Even foundational projects in the ecosystem like the Kernel itself now have policies around how to govern this, and will accept tasteful, correct contributions that have been authored with AI".
For a Linux distribution, getting away from AI code is going to be increasingly impossible it seems.
They also noted how none of this is in the recent Ubuntu 26.04 LTS release, these are just future plans for Ubuntu 26.10 and beyond.
Article taken from GamingOnLinux.com.
About the author - Liam Dawe
I am the owner of GamingOnLinux. After discovering Linux back in the days of Mandrake in 2003, I constantly checked on the progress of Linux until Ubuntu appeared on the scene and it helped me to really love it. You can reach me easily by emailing GamingOnLinux directly. You can follow me personally on Mastodon [External Link].
See more from me
See more from me
Some you may have missed, popular articles from the last month:
You can also find comments for this article on social media: All posts need to follow our rules. Please hit the Report Flag icon on any post that breaks the rules or contains illegal / harmful content. Readers can also email us for any issues or concerns.
9 comments
I find always... uhm... interesting to see how some corporations need to run to the damage control room right after announcements (that they fully know are cause for concern), and they rush to "clarify" only after the fecal matter meets the bladed air blower.
But never before that.
But never before that.
4 Likes
Ah, snaps. PERFECT. Go for it. Woo! Go go go. Since I remove everything snap from every installation, I'm a-ok with that.
*ahem*
More seriously, if they stick to their word about opt-in, visible onboarding, and all that, there's little room to complain for. Sure, it's another piece in the "AI everywhere", but the option is likely to always have materialized somewhere; better have it with full human control.
At least, as long as they stick to their words, which is not granted.
*ahem*
More seriously, if they stick to their word about opt-in, visible onboarding, and all that, there's little room to complain for. Sure, it's another piece in the "AI everywhere", but the option is likely to always have materialized somewhere; better have it with full human control.
At least, as long as they stick to their words, which is not granted.
0 Likes
benstor214 7 hours ago
“Opt-in” and “easy to remove”…
Yes, I can see how these tidbits of information were impossible to incorporate in the original press release and had to be moved to a separate damage control “clarification”. They are just too big. /s
I agree with @syylk: it is silly to observe that these two words were left out of the original release. Not that it would make any difference to me, I won’t touch their distribution with a 10-foot-pole.
Last edited by benstor214 on 28 Apr 2026 at 10:28 am UTC
Yes, I can see how these tidbits of information were impossible to incorporate in the original press release and had to be moved to a separate damage control “clarification”. They are just too big. /s
I agree with @syylk: it is silly to observe that these two words were left out of the original release. Not that it would make any difference to me, I won’t touch their distribution with a 10-foot-pole.
Last edited by benstor214 on 28 Apr 2026 at 10:28 am UTC
2 Likes
I definitely see that board rooms are genuinely enthusiastic about genAI and want to see it adopted everywhere, usually for "efficiency gains" (and I'll put aside for another discussion that this almost always means, but is never explicit said, "firing staff").
So there's a massive disconnect between these higher-management decision-makers, and the people who actually use the product that the decision-makers influence. Even the slightest awareness from these board execs would have prevented the need for Seager's hurried clarifications. He genuinely must have thought - "whoa boy, people are gonna LOVE this!" and out comes the press release.
Then, suddenly, bafflement from Seager/Canoncial, and damage control.
I'd say it's embarrassing, but that makes it sound like a little "whoopsadasie, sorry about that". Instead, this is deeply disrespectful. They're making decisions about a well-loved project, but without any awareness of a) the people that use it, or b) the complete shitstorm that MS went through just a handful of months ago by doing a very, very similar thing (yeah, yeah, it's local models, blah blah).
And I know this... that they have no awareness... because otherwise these simple clarifications wouldn't have been necessary at all - they'd have been explicitly mentioned in the initial release.
So there's a massive disconnect between these higher-management decision-makers, and the people who actually use the product that the decision-makers influence. Even the slightest awareness from these board execs would have prevented the need for Seager's hurried clarifications. He genuinely must have thought - "whoa boy, people are gonna LOVE this!" and out comes the press release.
Then, suddenly, bafflement from Seager/Canoncial, and damage control.
I'd say it's embarrassing, but that makes it sound like a little "whoopsadasie, sorry about that". Instead, this is deeply disrespectful. They're making decisions about a well-loved project, but without any awareness of a) the people that use it, or b) the complete shitstorm that MS went through just a handful of months ago by doing a very, very similar thing (yeah, yeah, it's local models, blah blah).
And I know this... that they have no awareness... because otherwise these simple clarifications wouldn't have been necessary at all - they'd have been explicitly mentioned in the initial release.
3 Likes
Are Canonical and Mozilla the same place? LOL it seems like they are doing the exact same things step by step. And by "same exact things" I mean "stupid dumb bullshit nobody asked for but is being forced down our throats".
1 Likes
Kimyrielle 3 hours ago
Their first post was clear enough - local, optional and easy to shut off. The sad part is really where they thought their second post would stop the rage coming from the same people who already didn't read the first. It only takes "AI" somewhere in the first line to make these people go reach for the pitchforks. Objectivity and reason has left this discussion a while ago.
5 Likes
Because of the size of most LLMs, we simply couldn’t ship them in the installer anyway, so opting out at first run is simple: they just won’t be there.I was also thinking this yesterday. Especially if they’re going to support any language other than English, including the models in the installer directly would create a huge amount of bloat. It wouldn’t have made any sense for it to be opt-out.
0 Likes
You can fit a small local model in 500MB, although naturally its accuracy will suffer.
For me, I only want to run models using my own choices of software. For instance, I would not be happy with Ubuntu or some corporation pushing some agentic features alongside the LLM. The agentic side is where A.I. gets into mischief - It exposes security holes, e.g. prompt attacks which can affect backend systems.
Also, agentic features enable Big Tech corporations (Microsoft, Google, OpenAI) to conceal telemetry and other self-serving capabilities (backdoors, etc). Naturally, any built-in agentic capabilities would be the first place hackers will be aiming to exploit.
For me, I only want to run models using my own choices of software. For instance, I would not be happy with Ubuntu or some corporation pushing some agentic features alongside the LLM. The agentic side is where A.I. gets into mischief - It exposes security holes, e.g. prompt attacks which can affect backend systems.
Also, agentic features enable Big Tech corporations (Microsoft, Google, OpenAI) to conceal telemetry and other self-serving capabilities (backdoors, etc). Naturally, any built-in agentic capabilities would be the first place hackers will be aiming to exploit.
0 Likes
Quoting: g000hYou can fit a small local model in 500MB, although naturally its accuracy will suffer.For camera focus, which is one of the features evoked by Seager, Canon, Sony, Nikon, Olympus, Panasonic and Pentax had integrated neural network trained using Deep Learning for autofocus in their DSLR and Miroless for around 15 years. This is notably how they greatly increased the number of species on which focusing using the "Focus on eyes" feature works.
For me, I only want to run models using my own choices of software. For instance, I would not be happy with Ubuntu or some corporation pushing some agentic features alongside the LLM. The agentic side is where A.I. gets into mischief - It exposes security holes, e.g. prompt attacks which can affect backend systems.
Also, agentic features enable Big Tech corporations (Microsoft, Google, OpenAI) to conceal telemetry and other self-serving capabilities (backdoors, etc). Naturally, any built-in agentic capabilities would be the first place hackers will be aiming to exploit.
So "AI powered" camera focus can use very small models depending of what you call AI powered.
0 Likes
Patreon
PayPal
How to setup OpenMW for modern Morrowind on Linux / SteamOS and Steam Deck
How to install Hollow Knight: Silksong mods on Linux, SteamOS and Steam Deck