A Promise, Quietly Erased
Alexander Hanff, a prominent privacy researcher, didn’t mince words when he flagged what Chrome had been doing in the background: silently installing a 4GB AI model on devices without user consent. His assessment was blunt — the practice may violate EU law. That’s not a minor footnote. That’s a researcher with standing calling out a potential legal breach by one of the most widely used browsers on the planet.
As someone who builds bots for a living, I spend a lot of time thinking about what runs on a device, what phones home, and what users actually agreed to. So when I read that Chrome quietly removed the line stating “Chrome can use AI models that run directly on your device without sending your data to Google servers” — a line that disappeared in Chrome 148.0 — I didn’t just raise an eyebrow. I started auditing my own assumptions.
What Actually Changed
Let’s be precise about what happened here, because the details matter more than the headline.
- Chrome previously made an explicit privacy claim: on-device AI meant your data stayed on your device.
- That claim has been removed from Chrome’s documentation.
- Chrome is now installing AI models — reportedly a 4GB package — on user devices without asking.
- New AI features are turned on by default after auto-updates, not opt-in.
- The change has triggered privacy concerns and, according to Hanff, potential scrutiny under EU law.
Removing a privacy promise isn’t the same as never making one. Google made a specific claim, users and developers made decisions based on that claim, and now the claim is gone. That’s not a product update. That’s a trust problem.
Why Bot Builders Should Care
If you’re building on top of Chrome-based infrastructure — and a lot of us are, whether through Puppeteer, Chrome extensions, or Chromium-based automation — this matters to your architecture decisions right now.
When I build a bot that interacts with a browser environment, I’m making assumptions about what that environment does with data. If the browser itself is running a 4GB AI model in the background and the data-handling terms have quietly shifted, those assumptions need revisiting. Your users trust your bot. Your bot runs inside Chrome. Chrome’s behavior is now less transparent than it was six months ago.
This is also a signal about how AI features are being deployed at the platform level. Not through clear opt-in flows. Not through settings menus most users will find. Through silent downloads and default-on toggles buried in auto-updates. For anyone building products where user trust is a core feature — and if you’re building bots, it absolutely should be — this is a pattern worth watching closely.
The Opt-Out Problem
One of the more frustrating details to surface in community discussions is that after Chrome auto-updates, new AI features are switched on by default. That means the burden falls entirely on users to discover, understand, and disable something they never agreed to enable.
Default-on is a deliberate design choice. It maximizes adoption numbers and minimizes friction for Google’s AI rollout. But it also means millions of people are running a 4GB model they didn’t ask for, on hardware they own, under terms that used to include a privacy guarantee that no longer exists.
For developers building user-facing tools, this creates a real tension. You can write solid privacy policies for your own product, but you can’t fully control what the underlying browser is doing. That gap is getting wider.
What to Do Right Now
If you’re building bots or browser-based tools and want to stay ahead of this, here are a few practical steps worth taking today.
- Check Chrome’s current AI settings in your own browser under
chrome://settings/aiand see what’s enabled by default. - Review whether your bot architecture has any dependency on Chrome’s built-in AI features — and whether that changes your data handling obligations.
- If you’re shipping tools to EU users, flag this development to your legal or compliance contacts. Hanff’s concern about EU law isn’t hypothetical.
- Consider documenting in your own product’s privacy materials that browser-level behavior is outside your control, and link users to Chrome’s settings.
Google may clarify its position, update its documentation, or roll back some of these defaults under regulatory pressure. But the fact that a clear privacy commitment was removed without announcement tells you something about how much weight that commitment carried in the first place. Build your architecture accordingly.
🕒 Published: