· via Hacker News – Front Page (hnrss.org)
Firefox 155 retest: AI switch stays off, telemetry and ads keep running
An independent Wireshark retest of Firefox 155 confirms the AI Controls toggle disables generative features while telemetry and ad traffic continue — and most users never discover the switch at all.

An independent packet-capture test of Firefox 155 concludes that the browser's AI on/off switch does exactly what it promises and nothing more: generative AI features are disabled, but telemetry and advertising connections carry on as before. The retest was published on September 2 by Marius Quabeck, whose earlier measurements of Firefox 148 found the same pattern, and the write-up climbed to the front page of Hacker News.
How the retest was run
According to Quabeck's post, Firefox 155.0 — released September 1 — was installed as a DEB package from Mozilla's official repository on a 2013 MacBook Pro (the A1502) running a clean copy of Ubuntu 24.04.3. Wireshark 4.2.2 captured all traffic, TLS was decrypted via the SSLKEYLOGFILE environment variable, and every step was performed by hand with mouse and keyboard in a real desktop session: no WebDriver, no remote debugging, no prepared profile.
Two factory-fresh profiles formed the test series. The A runs covered a cold start, setup, idle time and a fixed browsing scenario in the default state; the B runs repeated those steps with "Block AI Enhancements" enabled, plus a restart of the blocked profile. Crucially, both cold starts happened before anyone touched the switch, closing a methodological gap from the first test, where one-time first-run connections could masquerade as a blocking effect.
There is measurable good news. On first launch, Firefox 155 contacted 14 TLS hosts and produced 70 DNS queries, against 25 hosts and 76 queries in the March test of the older version. Quabeck credits that improvement to the browser generally getting quieter, not to the switch — and cautions that comparing a later nine-host restart against the 14-host first launch repeats the warm-profile fallacy, since the drop is a cache effect.
Inside the blocked profile, Firefox 155 marks generative AI features as blocked, and the control's scope has grown to include additional items such as Smart Window. The post's own section headings are blunt about the rest of the picture: "Telemetry does not understand 'blocked'" and "Advertising continues, and Temu says hello." Quabeck is explicit that the switch never claimed to disable telemetry or advertising; his argument is that many users believed it did, and that Mozilla should either broaden the existing control or add a second one beside it.
The CEO's essay hangs on the numbers
The retest lands in a changed climate. On August 25, Mozilla CEO Anthony Enzor-DeMeo published an essay titled "Trust is earned. Consent is asked. AI skipped both," recounting his first day in the job in December 2025, the backlash that followed the AI rollout — "It was a gut punch. Frankly, I deserved it" — and his decision to prioritise a switch that turns all AI features off. He prefers the name AI Controls over "kill switch."
The essay devotes a paragraph to the aftermath: "Almost nobody used it." In the months after release, roughly one percent of users switched off all AI, and about three percent disabled individual features. Enzor-DeMeo reads this as evidence that AI has no fixed definition — most users kept translation, an older form of machine learning, while rejecting generative features — and cites Pew Research figures showing about half of US adults use AI chatbots while 59 percent distrust US companies to develop AI responsibly. His closing line: "The most important AI feature may be the ability to turn it off."
Quabeck reads the essay as unusually self-critical for Mozilla, but also as public relations that converts a concession to protesting users into proof of the company's closeness to them — all resting on the very usage numbers the retest calls into question.
A switch most users cannot find
The CEO promoted the essay in a LinkedIn post that gathered 537 reactions and 53 comments within a week. Mozilla product manager Olivia O'Hara installed Firefox from scratch and reported that the onboarding flow never mentions AI features at all, let alone a switch; users must open the hamburger menu, enter Settings and select a dedicated tab. "It feels buried," she wrote, suggesting the one- and three-percent figures may be artificially low. Jed Wheeler added that low usage here simply signals poor design. The CEO replied that the switch sits at Settings, then AI Controls — the exact route O'Hara had just described as buried.
Another commenter, Andres Servin, drew the distinction at the heart of the debate: "optional" and "controlled" are fundamentally different, and users prefer a door that is locked by default to one that can be locked on request. Quabeck also notes that his earlier article was discussed inside Mozilla's Slack and labelled a "hit piece," even as it appeared on the cover of the US edition of Linux Magazine, issue 309, in August 2026.
Why it matters
Firefox sells itself as the privacy browser, and its leadership now frames the off switch as perhaps its most important AI feature. If that switch leaves telemetry and advertising running — and if the browser never tells new users it exists — then the "choice and agency" argument rests on a control most people never flip, and the headline usage percentages measure discoverability as much as demand. The practical takeaway for users is simple: the AI switch is not a privacy switch, and silencing telemetry and ad tracking requires digging into separate settings.
- #firefox
- #mozilla
- #privacy
- #telemetry
- #browsers