Tech

Google Chrome's 4GB AI model has been installed since 2024, but a recent settings update has reignited privacy concerns over data locality

Critics argue the shift to an opt-out default for on-device processing, coupled with ambiguous communication, undermines user trust in Google's AI strategy

Author
Owen Mercer
Markets and Finance Editor
Published
Draft
Source: Ars Technica · original
Chrome's 4GB AI model isn't new, but you're not wrong for being confused
Tech giant maintains data remains on-device despite removing explicit assurances from user interface

Google Chrome has been automatically downloading a 4GB Gemini Nano artificial intelligence model for on-device processing since 2024, yet users are only recently noticing the associated storage consumption. Although Google states the model size has remained constant and data is processed locally, the company recently removed a settings stipulation clarifying that on-device AI data does not send to Google servers. This change has drawn sharp criticism from observers who argue the move exacerbates existing privacy concerns.

The controversy centres on a recent update to Chrome version 148, which altered the label for the local AI toggle in the System settings. Google removed the specific text previously stating that on-device AI data would not be transmitted to Google's servers. A spokesperson for the company stated this wording change was made earlier in 2026 to ensure clarity regarding how AI works on the web, asserting that data passed to the model is still processed solely on the device.

Despite Google's assurance that the technical handling of data has not changed, the removal of the explicit stipulation has been interpreted by critics as a shift towards an opt-out default approach. This strategy, combined with confusing communication, raises questions about transparency as users increasingly seek to avoid AI features. The lack of explicit consent mechanisms at the point of installation has led to speculation that Google is relying on defaults to integrate these capabilities without asking for permission.

Google confirmed to Ars Technica that the 4GB size of the Gemini Nano model has been consistent since its debut two years ago. While local AI capabilities were announced in 2024 to power features like Help Me Write and tab organisation, the models were never rolled out universally to all installs. The company explains that the presence of the model depends on multifaceted factors, including hardware, account features, and whether a user has visited a website using Google's on-device Gemini API.

The storage impact of the model has become more visible to users as browser usage patterns evolve. A fresh Chrome installation typically consumes 6–8GB of storage initially and can grow to occupy 10 times that amount over a few months due to cache and extension data. While the 4GB AI model is a smaller portion of total browser usage, the recent removal of clarifying text in the settings has highlighted the issue for those attempting to manage their device resources.

Users retain the ability to disable local AI features via a toggle under the System tab, which will remove the model and prevent it from redownloading. However, the broader implication of Google's business model, which relies heavily on defaults, suggests that the company expects users to accept these features unless they actively intervene. As the backlash against AI grows, the discrepancy between Google's assurance of local processing and the lack of transparent user consent mechanisms remains a focal point for scrutiny.

Continue reading

More from Tech

Read next: Apple to roll out manual EQ controls for AirPods in iOS 27 update
Read next: Apple rolls out visionOS 27, integrating AI-driven Siri into Vision Pro headset
Read next: Apple Overhauls Siri with Google Gemini Partnership and Standalone App at WWDC 2026