Is Firefox AI Private? Mozilla's VP of Product on AI Opt-Out Controls, Smart Window, and Having a Seat at the Table

AI is everywhere — including Firefox. But unlike most services, Mozilla allows you to turn it off completely.

I went into this interview a little biased; I've been a Firefox user and Mozilla fan for a long time. I've also been tinkering with AI here and there to see what value it can add to my workflows. Some tools are genuinely transformational, while others feel useless. So when I learned Mozilla was adding AI to Firefox, I was interested to see what value they thought it would add.

Here's an overview of Henry's conversation with Adam Fishman, Mozilla's VP of Product, along with some of my thoughts.

Mozilla's reason for engaging with AI

The teaser clip I chose was Adam explaining why Mozilla can't sit this one out:

Adam: "Mozilla's never sat on the sidelines, like never. Going back 25 years. It's not our position to sit on the sidelines, especially if big things are shaping the future of how people use the web.”

He expanded on this later in the interview:

Adam: “Regardless of how you feel about AI, I think one thing that's pretty clear is we're moving towards this new era of how people and computers interact with each other, and AI is going to have a really big impact on shaping access and the use of the Internet and information. And that has always been Mozilla's focus.”

If the companies shaping AI in browsers are primarily Google and Microsoft, having Mozilla at that table matters. As the developers of Gecko, one of the only remaining independent browser engines, them engaging with AI thoughtfully is exactly what I want to see.

AI Features vs. Advocacy

Henry had a great question: How much of Mozilla’s new AI is about adding features to Firefox, and how much is about advocacy?

Adam touched on something I hadn’t thought of:

Adam: "If you're a very anti-AI person, and you're like ‘No AI anywhere!’ What if you were also blind, and you needed that information because most people don't add their own alt text to images? That is a genuinely useful feature to a large swath of the population."

It's a good reminder that "AI" isn't one thing. An on-device model quietly generating alt text for images that would otherwise be invisible to a screen reader is a very different proposition than a data-hungry cloud model. The use case Adam is describing has real, tangible value for people who are often an afterthought in how software gets built.

On-device AI is the way to go

Adam: “We have an on-device AI model in Firefox. So that means it's not hitting a cloud server. It's not sending data anywhere else. It's literally doing this on your device.”

Here's where I land personally: if it's on-device and open source, I have zero issues with it. In fact, I’m excited about it. The productivity potential alone is genuinely compelling. What I don't want is my data leaving my machine to train someone else's model.

I think Firefox's Smart Window feature is a good approach. It stores context locally, doesn't sync to Mozilla's servers, and you can delete the models from your device entirely.

The global off switch matters

This is the kicker. The opt-out is real. It’s not buried, and it works.

Adam: “You can turn it all off if you want. And that is a feature that I have not seen any other browser do, is the ability to have a toggle that says, ‘get all this stuff out of my face, delete all the models that are locally on my machine, like gone. I don't want to be aware of it. I don't want to be exposed to it. I don't even want to see a marketing message that says you should use it.’ You can turn it all off. And then you can selectively reactivate stuff, if you want.”

In a world of enshittification, terrible design choices, and forced updates, Mozilla building what is effectively an AI kill switch is worth applauding.

The Anthropic security partnership

Anthropic’s security Red Team used Claude Opus to audit Firefox code:

Adam: "Opus found 22 security vulnerabilities in a span of two weeks, which is 30-40% of all of the vulnerabilities that we found the entire previous year."

They started submitting PRs, and Mozilla noticed quickly. They reached out to Anthropic, and have since been working closely with them to rapidly fix security vulnerabilities.

From a productivity and security standpoint, this is exactly the kind of AI application I want to see normalized. Not vibe coded apps riddled with bugs trying to replace human skill and judgment, but AI doing tedious, high-stakes work faster than any human team otherwise could.

What I'm watching going forward

I'm optimistic, but not uncritical. The features that ship on by default and the pace of the rollout are things worth monitoring. Open source and good intentions are a strong foundation, but implementation details matter, and it's still early.

I'm also curious to see how the on-device model ecosystem develops. The more capable local models get, the more useful these features become without requiring any privacy tradeoffs. Mozilla is in a good position to push that forward, and I hope they do.

Episode Sources

• Firefox: https://www.firefox.com
• Mozilla AI: https://www.mozilla.ai
• Mozilla Common Voice: https://commonvoice.mozilla.org

Written by

Share

Copied