Google Chrome has come under scrutiny after reports revealed that the browser has been automatically downloading a large AI model onto some users’ devices without clearly requesting permission.
Privacy researcher Alexander Hanff said he discovered the issue while running automated privacy audits through a Chrome profile that had never been actively used by a person. During the investigation, he found approximately 4GB of AI-related files stored inside a directory labeled “OptGuideOnDeviceModel,” including a file named weights.bin.
According to Hanff, the downloaded model is Gemini Nano, a compact large language model designed to operate directly on local devices rather than entirely through cloud servers.
Evidence gathered during the investigation reportedly showed Chrome downloading the multi-gigabyte package within roughly 14 minutes on April 24, 2026. Hanff claimed the installation occurred without a visible consent request, pop-up notification, or dedicated opt-in setting presented to the user.
He also reported that the files automatically reappeared after deletion when the browser restarted. Similar behavior was later described by users across multiple operating systems, including Windows, macOS, and Linux.
The findings sparked additional discussion regarding how the browser is using the locally stored AI model. In Google Chrome version 147, users can see an “AI Mode” shortcut in the browser’s address bar, which some may assume relies on the locally installed model for generating responses.
However, Hanff’s research suggested that most AI-related queries are still processed through Google’s cloud infrastructure rather than directly through the on-device model. According to the investigation, Gemini Nano is primarily connected to smaller browser features accessible through contextual menus and other limited interface tools.
Fact-checking publication Snopes later reviewed the claims and reported finding similar AI model files on several staff devices running both Windows and macOS. Google reportedly told the publication that an opt-out setting for the feature began rolling out earlier in 2026, though the option was not yet universally available to all users.
The incident has intensified broader concerns surrounding silent software deployments, background data collection, and transparency in AI-powered platforms. Critics argue that users should receive clearer disclosures when large AI systems are added to their devices automatically.
Hanff also raised legal concerns, suggesting the practice could potentially conflict with European Union privacy regulations, including rules under the ePrivacy Directive and GDPR transparency requirements. Those claims, however, have not been tested in court.
Beyond privacy concerns, Hanff estimated that distributing a 4GB AI package across potentially billions of devices could create a substantial environmental impact due to the energy consumption involved in global data transfers.
Cybersecurity researchers also noted similarities to earlier reports involving Anthropic and its Claude Desktop software, which allegedly installed browser integration components on Chromium-based browsers without prominent disclosure.
The situation reflects growing anxiety around AI-driven privacy and security risks as technology companies rapidly integrate artificial intelligence into mainstream consumer software. Industry observers warn that insufficient transparency around these deployments could further weaken public trust in digital platforms and online services.
