Changes to Chrome's On-Device AI Privacy Statement Stir Controversy; Google Insists Processing Remains On-Device
Google has removed the phrase "does not send data to Google servers" from Chrome's on-device AI privacy description, raising user concerns. Google claims there has been no change in processing methods.
Key Phrase Removed from Chrome’s Settings
A change to the privacy description for Google Chrome’s on-device AI feature has sparked debate among users. The issue stems from the removal of the phrase “does not send data to Google servers” from the description of the “On-Device AI” section in Chrome’s “System” settings.
The previous description read: “To enhance features like fraud detection, Chrome can use AI models that run directly on your device. This does not send data to Google servers. Turning this off may cause these features to not work.” In the current version, the portion stating “This does not send data to Google servers” has simply been removed.
Questions from Privacy Advocates
Alexander Hanff, a privacy advocate who noticed the change, raised questions about its intent. Hanff asked, “Why has the phrase ‘does not send data to Google servers’ been removed from the description of on-device AI? Was the previous statement inaccurate? Has the architecture changed? Or was it withdrawn on legal advice because the claim could not be defended?”
This observation coincided with the rollout of Chrome’s “Prompt API,” which facilitates programmatic interaction between web pages and the browser’s built-in AI model. The introduction of this API also brought attention to the fact that Chrome had silently been downloading Google’s 4GB Nano model to users’ devices. These developments have fueled speculation that Google might be preparing to collect prompts and responses executed on-device, potentially marking a step back in terms of privacy.
Google Insists “No Change in Processing Method”
In response to these concerns, a Google spokesperson issued a statement: “This does not change how Chrome’s on-device AI processes data. Data passed to the model is processed on the device.” Google emphasized that the processing remains confined to users’ devices.
In reality, Chrome has been setting up this Nano model on devices since it was first implemented in Chrome 126 in 2024 as a preview for early adopters of local AI. Although Google has not yet made downloading and storing the model an opt-in process, it introduced a method earlier this year allowing users to disable and delete the large model. Google stated, “Since 2024, we have been offering the lightweight Gemini Nano model in Chrome.”
Background and Future Implications
This controversy highlights the challenges technology companies face in maintaining transparency and trust when offering on-device AI functionality. While performing processing locally is a strong point for privacy protection, insufficient or altered explanations can erode user trust.
Amid existing concerns about AI models occupying significant device storage, the perception of a rollback in privacy assurances raises issues for both user experience and trust. While Google has clarified that no technical changes have been made, the “poor timing” of this incident underscores the need for more cautious and transparent communication in the company’s future on-device AI strategies.
FAQ
Q: What is Google Chrome’s On-Device AI?
A: It is an AI model integrated into Chrome to enhance features like fraud detection. Its key advantage is performing data processing directly on the user’s device, generally ensuring that information is not sent to external servers. This allows users to leverage AI functionality while protecting their privacy.
Q: Why is the change in the privacy statement a concern?
A: The removal of the previously explicit assurance that “data will not be sent to Google servers” has led to concerns among users that their data might now be transmitted externally, or that the processing architecture has changed. Google has stated that there has been no change in the way data is processed.
Q: Can users delete the AI model from their Chrome browser?
A: Yes, they can. Google has introduced a function allowing users to disable and remove the AI model from devices. However, the model is currently downloaded automatically.
Comments