Trust is earned. Consent is asked. AI skipped both.

My first day as CEO was humbling.

I had spent the previous year leading Firefox and working with the team to turn the product around. I was excited about what we could do next.

In the media interviews before my announcement, I talked about evolving Firefox and bringing AI into the browser. The AI features we were building were optional, and I said as much. If you were to ask me what could go wrong, mentioning the word AI would have been last on my list.

Then the backlash came. It was a gut punch. Frankly, I deserved it.

I said the right words and still missed the message. We value AI as a tool, and it should always be a choice for our users.

So one of my first decisions as Mozilla's CEO wasn't to launch an AI feature; it was to build the switch that turns them all off.

The infamous ‘Kill Switch’, or as I like to say, AI Controls, lets people turn off some or all of the AI features in Firefox. It was an important feature we expedited post-announcement, not just because of the backlash, but because I, too, was getting annoyed seeing AI pop up everywhere.

I was proud that Mozilla was in a position where we could let users decide “how they AI”. Then we released it.

Almost nobody used it.

In the months following the launch of AI Controls within Firefox, about 1% of users turned AI off completely, and 3% turned off specific AI features. As an example, most people found value in things like translation, which is machine learning. Features that felt closer to generative AI were more likely to be turned off.

One of the lessons I took away from this experience is that AI, while being a trigger word for many, has no agreed-upon definition. Machine learning (ML) has been a rudimentary form of artificial intelligence for decades. Firefox features like Translations and internal tooling such as bug triaging rely on it. The newer type of artificial intelligence, generative learning or generative AI, is actually a subset of ML. The key difference is that gen AI creates new, often derivative content. I believe, in part, that the act of creation is what is polarizing to many.

It would be another mistake to read that as people loving AI. They don't. Pew found this year that about half of US adults now use AI chatbots, up from a third two years ago, and that 59% are not confident US companies will develop and use AI responsibly.

In a world where companies are forcing AI on users, the most compelling thing Firefox can do is stick with what we do best: give users choice and agency.

Firefox users taught me something incredibly valuable: that having a way out, an option, is the most important thing. I've spent the past year talking about AI's trust problem; turns out I was only half right. Trust matters, but it starts with control. People need to understand what their technology is doing and have a clear way to decide what works for them.

The most important AI feature may be the ability to turn it off.