Firefox 148 Releases with Promised AI Killswitch Feature

Firefox version 148 has released, bringing with it the AI killswitch feature that was promised, allowing users to disable all AI features from a single switch.

The switch is called “Block AI Enhancements” in the settings, and Mozilla says you can use it if you “don’t want to use AI features from Firefox at all.”

You can also individually enable or disable specific features. The features you can pick and choose from for now are translations, alt text in PDFs, AI-enhanced tab groups, link previews, and the AI chat sidebar.

AI features are blocked by default and must be opted into.

The new panel helpfully labels on-device AI features and those that send data off-device, such as the chat sidebar. To me, this is a great move for transparency, as it might have been unclear before whether a particular feature was on-device or was actively sending data off.

The feature, announced months ago, comes amid a huge AI-focused push at Mozilla, with their flashy State of Mozilla annual report being completely comprised of AI and nothing else for Firefox users to look forward to.

Mozilla is investing in over 50 AI companies to form what they call a “rebel alliance” against “big tech” AI companies like OpenAI. There’s not any mention of a focus on local-only AI or technologies that can enhance the privacy of AI like TEEs or homomorphic encryption. It’s mostly vague claims about these companies being better without much technical information to back it up.

Mozilla are focusing their “~$1.4B in reserves, our brand and our people power“ toward their idea of “responsible“ AI. One wonders what could be accomplished with that much money, especially considering Mozilla’s older projects like Rust, a memory-safe programming language that stops entire classes of exploits, and Servo, a rust-based browser engine that was originally meant to replace Firefox’s Gecko engine, that are now under the stewardship of other entities.

It’s unclear what Mozilla plans to expand their AI ventures into next. The hot new thing is agentic AI, which can perform actions on your behalf like sending emails or even shopping.

Google Chrome has already added an “auto browse” feature. In their security blog, Google highlights the challenges they face in trying to make agentic AI secure in the face of prompt injection attacks.

Trail of Bits, a cybersecurity consulting firm, has made multiple blog posts exposing the insecurity in currently-implemented agentic AI browsers. Mozilla needs to be careful about their future steps, as AI can be dangerous for their users’ data and security. They need to focus on local-only and non-agentic AI features as much as possible. I think this new AI killswitch is a great step in the right direction for user choice.