Google Chrome is downloading a large artificial intelligence model to user devices without explicit permission. Researcher Alexander Hanff claims the browser silently installs a roughly 4GB on-device AI model. This behavior occurs without notice or consent from the user. The download happens proactively rather than in response to a specific user action.
Installed model based on lightweight Gemini Nano architecture
The installed model is based on Google’s lightweight Gemini Nano architecture. It saves a file named weights.bin to the local disk. Hanff verified this behavior during a controlled test on macOS. He used a fresh Chrome profile to monitor filesystem events. The download process took just over fourteen minutes to complete.
Hanff argues that this practice may violate EU regulations. He cites the ePrivacy Directive and GDPR as potential legal frameworks. The claims regarding illegality have not been tested in court. Hanff also estimates the environmental impact of distributing this model. He suggests the energy cost could reach 24 GWh for 100 million devices.
These environmental estimates depend on assumptions about scale and energy mix. The source notes these assumptions may be dubious. The analysis was published by Tom’s Hardware and That Privacy Guy. The incident highlights growing concerns about browser transparency and data usage.
Google has not confirmed the launch window or official stance on this feature. The vendor has not addressed the specific privacy concerns raised by the researcher. The behavior remains unconfirmed by official Google channels.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.