The Next Web reports that Mozilla built an AI "kill switch" into Firefox, and only 1% of users have flipped it fully off while another 3% disabled some AI features selectively. CEO Anthony Enzor-DeMeo told the outlet: "Our community was pretty vocal... At its core, we want to listen to our users. It was honestly on the roadmap, but I expedited it, given the community feedback." Mozilla launched Smart Window in beta - a browsing mode offering three hosted AI backends (Google Gemini Flash Lite, Alibaba Cloud Qwen3-235B-A22B, and OpenAI GPT-OSS 120B) plus a custom bring-your-own endpoint - and The Next Web reports Mozilla does not use chat data to train models and filters sensitive information. Mozilla also launched a free built-in browser VPN with 1.5 million signups and roughly 800,000 active users, and plans a fall redesign codenamed Project Nova with up to 9% faster page loads, per The Next Web.
What happened
The Next Web reports that Mozilla added an AI "kill switch" to Firefox, and only 1% of users have flipped it fully off, with a further 3% turning off some AI features selectively. The Next Web quotes Mozilla CEO Anthony Enzor-DeMeo: "Our community was pretty vocal, especially during the CEO announcement, that not everyone wanted AI. At its core, we want to listen to our users. It was honestly on the roadmap, but I expedited it, given the community feedback." Mozilla launched Smart Window in beta on Firefox 150+, offering three hosted AI backends - Fast (Google Gemini Flash Lite), Flexible (Alibaba Cloud Qwen3-235B-A22B), and Personal (OpenAI GPT-OSS 120B) - plus a custom bring-your-own endpoint for self-hosted models, per OMG Ubuntu's hands-on. The Next Web reports Mozilla does not use chat data to train models and that it automatically filters out sensitive information.
VPN and market context
The Next Web reports the built-in browser VPN launched last month with 1.5 million signups and roughly 800,000 active users; the VPN encrypts only browser traffic, not system-wide connections. PCWorld reports Firefox holds roughly 3.45% of the US browser market; The Next Web puts global share at just over 2%. Chrome holds roughly 70% and Safari roughly 16% by those measures.
Project Nova redesign
The Next Web reports a fall redesign codenamed Project Nova, slated for September or October, including up to 9% faster page loads, compact mode, rounded UI elements, and AI-powered tab grouping. PCWorld's April 2026 interview with Firefox product head Ajit Varma confirmed Project Nova is the browser's first major interface overhaul in about six years, with emphasis on customization and privacy-tool integration including VPN, Relay email masking, and Monitor.
Industry context
Browser vendors are managing AI adoption pressure against vocal user resistance. The Register reported in February 2026 that Firefox 148 shipped the kill switch, drawing comparisons to Vivaldi's deliberate pull-back from generative features. Mozilla's three-tier model (hosted Gemini, Qwen3, and OpenAI backends, plus BYO) avoids locking users into a single AI provider - a pattern Varma described in his April interview as a deliberate choice to avoid becoming "a Copilot app" and to preserve an open internet.
What to watch
- •Smart Window model-choice distribution as beta expands beyond US and Canada.
- •Whether BYO-endpoint support extends to enterprise Firefox, given Mozilla's reported push into enterprise deployments.
- •VPN retention over time (800K active vs 1.5M signups implies roughly 53% retention) and whether the browser-only usage cap drives paid upgrades.
Scoring Rationale #
Browser-level BYO AI with opt-out telemetry is relevant to AI deployment practitioners: Mozilla's kill-switch data (1% full opt-out, 3% partial) offers a rare public measurement of AI feature rejection in a major consumer app, and Smart Window's three-model architecture confirms the multi-backend browser AI pattern. 6.6 reflects notable but not major significance - single-source primary reporting on product metrics for a sub-4% market share browser, corroborated on technical details by hands-on reviews and a PCWorld interview.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.