Google Chrome Silently Installing AI Model on Devices, Investigation Finds

Google Chrome has been quietly installing a 4-gigabyte artificial intelligence model on users’ computers without explicit consent, according to a recent investigation that has raised concerns about data privacy and storage usage.

The model, called Google Gemini Nano, was discovered on multiple devices running Chrome on both MacOS and Windows systems. The AI model is stored in a large file called “weights.bin” within a folder labeled “OptGuideOnDeviceModel” in Google Chrome’s user data directory.

Alexander Hanff, a computer scientist and privacy lawyer who runs ThatPrivacyGuy.com, first reported the finding on May 4, 2026. Hanff documented how Chrome was downloading the substantial AI model without notifying users or seeking permission. His claims quickly gained traction across social media platforms including Reddit, Threads, and X.

“I fully intend to file criminal charges against Google for this,” Hanff told Snopes in an email, arguing that Google’s actions violated European digital privacy laws that require companies to give consumers the choice to opt in to data collection rather than forcing them to opt out.

When contacted about the issue, Google confirmed the presence of the AI model but defended its implementation, stating: “We’ve offered Gemini Nano for Chrome since 2024 as a lightweight, on-device model. It powers important security capabilities like scam detection and developer APIs without sending your data to the cloud.”

The company acknowledged that the model “requires some local space on the desktop” but claimed the model would “automatically uninstall if the device is low on resources.” Google also noted that in February 2026, they began rolling out settings that allow users to disable and remove the model directly in Chrome.

Independent verification conducted by Snopes found the AI model present on three out of six staff computers tested – two running MacOS and one on Windows. This suggests the model isn’t installed universally but depends on various factors, including a device’s technical specifications and available storage space. According to Google’s developer documentation, the model is not installed on mobile devices due to compatibility issues.

The Gemini Nano model appears to support several AI features within Chrome, including the “Help me write” function and on-device scam detection. However, the unexpected storage consumption – approximately 4 gigabytes – has raised concerns among users who weren’t aware of this background installation.

Multiple social media posts and blog articles have claimed that Chrome automatically re-downloads the file if users attempt to delete it directly, though this behavior has not been independently verified by Snopes at the time of reporting.

For users concerned about the presence of this AI model on their devices, there is a way to check and remove it. Users can type “chrome://on-device-internals” in their address bar and enable debugging tools to view the model status. If the model is installed, users can navigate to Chrome’s settings page, click the “System” tab, and disable the “On-System AI” option, which should remove the model and prevent it from downloading again.

Hanff noted in his correspondence with Snopes that this setting wasn’t available to him, suggesting the opt-out option may still be rolling out to users globally. As an alternative, he reported success by manually disabling Chrome’s AI features through the “chrome://flags” page, specifically by turning off the “optimization-guide-on-device-model” flag.

The discovery raises important questions about transparency in software updates and data practices. While on-device AI processing can enhance privacy by keeping data local rather than sending it to cloud servers, the lack of clear notification about such substantial downloads has sparked debate about user consent and control over their own devices.

Industry observers note that as AI features become increasingly integrated into everyday applications, the balance between convenience, performance, and transparent user choice will remain a critical consideration for technology companies and privacy advocates alike.