Firefox dev clarifies there will be an AI 'kill switch'

I don't think more chatbots and such will make Firefox better, but there is one recent AI feature of Firefox that although I don't use it I think is very good.

The build in translation capability. I had to download all kinds of sketchy extensions to get that to work right for my uncle and now I can just flip a setting.

Quoting: LoudTechieI don't think more chatbots and such will make Firefox better, but there is one recent AI feature of Firefox that although I don't use it I think is very good.

The build in translation capability. I had to download all kinds of sketchy extensions to get that to work right for my uncle and now I can just flip a setting.

As far as I'm aware, and I am no expert so I could easily be wrong, but translation is generally Machine Learning (ML), rather than Large Language Model (LLM), which the chatbots and generative AI are built on. It does fall under the "AI" umbrella, but doesn't have the same issues as LLMs such as rampant plagiarism, enormous energy use, black box processing that cannot be audited, generating misinformation, etc. So therefore not really a big issue despite both being under the same umbrella.

Quoting: _wojtekIMHO the issue with Firefox "adopting AI" is way overblown. Yes, AI is not great (especially if it's done with speculative bubble and cash burning by big-tech) and it's questionable ethics wise but… sometimes it's usefull and people do use it…

Also what's the alternative? Crappy reskins of chromium? :/

Librewolf
Zen Browser
Waterfox

there you go, three alternatives that have no ai integrations whatsoever.

Calling Firefox a product multiple times makes my soul cringe.

Quoting: Nezchan

Quoting: LoudTechieI don't think more chatbots and such will make Firefox better, but there is one recent AI feature of Firefox that although I don't use it I think is very good.

The build in translation capability. I had to download all kinds of sketchy extensions to get that to work right for my uncle and now I can just flip a setting.

As far as I'm aware, and I am no expert so I could easily be wrong, but translation is generally Machine Learning (ML), rather than Large Language Model (LLM), which the chatbots and generative AI are built on. It does fall under the "AI" umbrella, but doesn't have the same issues as LLMs such as rampant plagiarism, enormous energy use, black box processing that cannot be audited, generating misinformation, etc. So therefore not really a big issue despite both being under the same umbrella.

Machine learning is the base category. LLM is a subcategory of machine learning.

My understanding of the training methods of Mozilla are insufficient to classify it, it as a LLM, but it certainly is an LM(language model).
I would guess based on their github and budget that it's not(LLM requires self-reinforcement learning, which costs a lot).

Copyright wise all translation is the creation of derivative works although many nations tend to have personal use exceptions(in this case this matters, because I'm talking about the act of translating the webpage not the act of training the model).

Like all ML it's trained on data produced by people and thus at least subject to copyright.
Whether or not they've permission from the copyright holders to train their model with it I'm uncertain.
I would guess they've, because it's public which data they use and they've yet to be sued into oblivion.

[It's the most auditable model I've encountered in quite some time, but you still can't usefully attach a debugger to it](https://github.com/mozilla/translations)
It's translations certainly aren't perfect, does mistranslated information count as misinformation?

Last edited by LoudTechie on 19 Dec 2025 at 3:49 pm UTC

Quoting: Lofty

Quoting: doragasuToo late, today I switched to LibreWolf after literally DECADES of loyalty, and everything is working great, I'm not going back unless things change A LOT.

Also about that killswitch, let me guess, it will be opt-out and will reset each time a update is installed, right?

btw read those settings above. They are currently enabled in Librewolf too.

@Lofty
This contradicts with @redneckdrow's observation (and @toru9999's too) in the other thread:

Quoting: redneckdrow

Quoting: toru9999about:config
browser.ml.enable -> false
problem solved

Just checked, this is the default in Librewolf.

I'm not able to confirm this setting at the moment. Have you looked into it?

I'm assuming that you can enable ml by copying a Firefox profile to Librewolf.

To play devil's advocate... Firefox incorporaing random AI features for whatever corporate funding / CEO schmoozing in addition to normie appeal, while also making it easy to disable AI, seems like a least worst case for Firefox in the current moment without plunging itself into further irrelevancy.

I for one am not keen on having to install a chromium based browser so I'm feeling pretty generous towards Firefox doing what it has to do to survive.

And for all the Stallmanites on here, there are obscure FF forks for you, WHICH Firefox's continued existence and survival (which may require some corporate AI BS) depend on, I might add.

Quoting: Caldathras

Quoting: Lofty

Quoting: doragasuToo late, today I switched to LibreWolf after literally DECADES of loyalty, and everything is working great, I'm not going back unless things change A LOT.

Also about that killswitch, let me guess, it will be opt-out and will reset each time a update is installed, right?

btw read those settings above. They are currently enabled in Librewolf too.

@Lofty
This contradicts with @redneckdrow's observation (and @toru9999's too) in the other thread:

Quoting: redneckdrow

Quoting: toru9999about:config
browser.ml.enable -> false
problem solved

Just checked, this is the default in Librewolf.

I'm not able to confirm this setting at the moment. Have you looked into it?

My version is the Flatpak, idk if that is any different from any other sources. Perhaps there was an update recently, i did check early yesterday.

thx

That's...better, I guess.

Ah well. Already switched to Floorp. Had your opportunity Mozilla, sorry

"You will have a clear way to turn AI features off. A real kill switch is coming in Q1 of 2026."

A lot of people have their eyes on them at the moment and will look at them in 2026 about this.
They have no room for more mistakes, they better have that "killswitch" to work perfectly without any flaws or ambiguity.
Because some AI features have already been implemented without that kind of clear switch and it's been like a headache to disable all of them through the about:config panel, and again check that after updates etc (not to mention that it's only for tech-savvy people, that's a problem itself).
So yeah, it's pretty easy to see how a lot of us have lose their trust, I don't blame them.

I agree with Liam on this: "but the fact that they're still pouring resources into AI is a problem". I just hope that AI bubble won't sink Mozilla entirely because of that.

Another thing that really bothers me about these marketing statements, is the blanket statements of calling everything "AI".
"AI" doesn't actually mean anything and could be anything from a local ML model that can, say, give names to colours based on the "training data" that could be the list of X11 colour names, to a deepfake slop machine that uses the power budget of a city.
When an announcement like this is made, with practically no actual information other than "ridiculous buzzword that's either a decade old technology that can run locally on your watch, or a privacy nightmare of a planet destroyer - who knows! Go shareholders!", how is anyone supposed to take it seriously and come away with any other interpretation than that they just don't care which it is, as long as "money might happen"?

Last edited by Cybolic on 20 Dec 2025 at 12:28 am UTC

Quoting: LoudTechie...
Like all ML it's trained on data produced by people and thus at least subject to copyright.
Whether or not they've permission from the copyright holders to train their model with it I'm uncertain.
I would guess they've, because it's public which data they use and they've yet to be sued into oblivion.

[It's the most auditable model I've encountered in quite some time, but you still can't usefully attach a debugger to it](https://github.com/mozilla/translations)
It's translations certainly aren't perfect, does mistranslated information count as misinformation?

Haven't got the time to search for the source, but I understood mozilla uses open datasets for their models, i.e. permission was granted or copyrights expired.

I find it amusing, by the way, that the single "ai" browser feature that is actually useful to me is rarely advertised as being such. Even the "anti-ai" vivaldi incudes it without it raising a single eyebrow.

Last edited by emphy on 20 Dec 2025 at 12:57 am UTC

Quoting: Bumadar

Quoting: CentrisToo late. Already got rid of FF.

Maybe I am of an older generation, but why this on/off view?
Mozilla says they going to do AI in the future somewhere, it's not clear exactly how or what this exactly means or how it will be incorporated but you drop it like a hot potato and suddenly overnight they are evil.
The world is many shades of gray, not just black and white, see how it onfolds and then take a decision.

Because "ai" was merely the proverbial straw?

For years now, mozilla has failed to truly address the issues the organisation has with corporate bloat and the addiction to their so-called ethical monetisations.

Imo, this recent carfuffle with fake intelligence is merely a symptom, not the problem.

Quoting: emphy

Quoting: LoudTechie...
Like all ML it's trained on data produced by people and thus at least subject to copyright.
Whether or not they've permission from the copyright holders to train their model with it I'm uncertain.
I would guess they've, because it's public which data they use and they've yet to be sued into oblivion.

[It's the most auditable model I've encountered in quite some time, but you still can't usefully attach a debugger to it](https://github.com/mozilla/translations)
It's translations certainly aren't perfect, does mistranslated information count as misinformation?

Haven't got the time to search for the source, but I understood mozilla uses open datasets for their models, i.e. permission was granted or copyrights expired.

I find it amusing, by the way, that the single "ai" browser feature that is actually useful to me is rarely advertised as being such. Even the "anti-ai" vivaldi incudes it without it raising a single eyebrow.

This could be my fault.
I treat AI as equivalent to ML, but they're different words.

Wikipedia defines AI as

[Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence](https://en.wikipedia.org/wiki/Artificial_intelligence)

and ML as

[Machine learning (ML) is a field of study in artificial intelligence concerned with the development and study of statistical algorithms that can learn from data and generalize to unseen data, and thus perform tasks without explicit instructions.](https://en.wikipedia.org/wiki/Machine_learning)

.

If we take these definitions.
Aimbots are AI by this definition, but is translation AI.
It's an application of Machine Learning, but that wasn't the question.
Do we typically associate translation with human intelligence and problem solving.
Is someone appear more intelligent when they can successfully translate a sentence or would we say that communication through a matching language is the intelligent part and thus using a matching a language to the one used toward you is AI.
I would argue the first, because people tend to think that people who use complicated text are more intelligent, meaning that understanding of language is more intelligent than the capacity to communicate. I disagree with my preceived majority, but that doesn't change that it's commonly associated with human intelligence.

Quoting: LoftyMy version is the Flatpak, idk if that is any different from any other sources. Perhaps there was an update recently, i did check early yesterday.

thx

I'm using the AppImage myself. browser.ml.enable does default to false but, for some reason, mine was set to true. Don't know if I wasn't paying attention and set it to true by mistake or if it shipped that way. The only thing I import is the bookmarks. I always manually configure the settings and plugins.