Chrome Quietly Removed Its Privacy Promise for On-Device AI – Here’s What to Do
If you use Chrome and have ever glanced at the browser’s privacy policy, you might have noticed a specific reassurance: on-device AI features process everything locally, and Google does not receive any data from that processing. That sentence is no longer there.
Multiple news outlets reported in early May 2026 that Google silently deleted the explicit privacy promise from its AI documentation. At the same time, Chrome has been automatically downloading a 4-gigabyte AI model to many computers—and, as some users found, putting it back even after deletion. Here is what changed, why it matters, and how to take control of your privacy.
What happened
The removed promise, according to coverage from Decrypt and GIGAZINE, was a line stating that “Google does not receive any data from on-device AI processing.” It appeared in the help pages for Chrome’s on-device AI features, which include things like smart text selection, tab grouping suggestions, and writing assistance. The page was updated sometime in late April or early May 2026, stripping that assurance.
In its place, the language now describes how the AI model works locally but no longer makes any claim about what data stays on your device. The change is subtle—if you are not comparing old and new versions side by side, you would likely miss it.
Meanwhile, Chrome has begun quietly installing a 4GB AI model on compatible machines. According to reports from Decrypt, even if you delete the model files, Chrome can redownload them on the next launch. The feature is rolling out gradually, meaning not every user sees it yet.
Why it matters
The core issue is trust. On-device AI is sold as a privacy advantage: your data never leaves your machine, so no server-side logging, no training on your inputs, no exposure to a cloud provider. By removing the explicit promise, Google leaves room for interpretation—and for potential future changes in how data is handled, even if no actual policy shift has been confirmed.
Google has not publicly explained why the sentence was removed. The company may have updated the wording for other reasons, such as legal precision, but without a clear statement, users are left to wonder. Given that the feature is being automatically installed and persistently reinstalled, the concern is less about current data handling and more about what the future holds.
What you can do
You have several options, depending on your comfort level and how much you rely on Chrome.
1. Check whether the AI model is installed.
In Chrome, go to chrome://settings/ai (the exact path may vary by version). Look for “On-device AI” or “AI features.” If the feature is on, you will see a toggle and likely a download size. You can disable it from there.
2. Turn off on-device AI entirely.
Under the same settings page, switch off any toggle related to “Help me write,” “Tab organizer,” or similar on-device AI tools. This should prevent the model from running, though it may still be downloaded in the background.
3. Delete the model and prevent reinstallation.
If you want to remove the downloaded model, you can find it in Chrome’s profile folder (locations differ by OS). However, as reports note, Chrome may re-download it. A more reliable approach is to disable the feature first, then delete the model files, and then monitor whether they reappear.
4. Switch to a different browser.
For users who want to avoid Google’s ecosystem entirely, browsers like Firefox, Brave, or Vivaldi offer strong privacy defaults without automatically downloading large AI models. Edge has its own on-device AI but with different privacy documentation. No alternative is perfect, but they offer more transparency around what is being installed.
If you do want to keep using Chrome’s on-device AI, you can mitigate risk by keeping the feature disabled except when you explicitly need it, and by reviewing the privacy policy periodically.
Broader implications
This incident is a reminder that browser privacy promises can change without a formal announcement. On-device processing is still generally safer than cloud-based AI, but the removal of a clear data-sharing denial introduces ambiguity. For now, the safest course is to assume that any data processed by Chrome’s AI tools could potentially be sent to Google at some point, even if that is not currently happening.
As with any software that downloads large models to your machine without asking, the best defense is to know what is installed and to decide for yourself whether you want it.
Sources
- Decrypt: “Chrome Deleted Its Own Privacy Promise for Sneaky On-Device AI” (May 7, 2026)
- Yahoo Tech: coverage of the same report (May 7, 2026)
- GIGAZINE: “Chrome removes claim that its ‘on-device AI’ does not send data to Google servers” (May 9, 2026)
- Decrypt: “Chrome Is Quietly Installing a 4GB AI Model on Your Computer—And Putting It Back If You Delete It” (May 6, 2026)