Google Chrome accused of silently installing Gemini Nano files
Summary
Computer scientist Alexander Hanff alleges that Google Chrome silently downloads a 4GB "weights.bin" file containing data for the Gemini Nano on-device large language model. This file, found in a hidden macOS Library folder after updating to Chrome version 148.0.7778.97, reportedly powers AI features like "help me write" and scam detection and reappears after deletion. Google confirmed Gemini Nano's presence in Chrome since 2024, emphasizing its lightweight, on-device nature for security and developer APIs without cloud data transfer. Hanff observed similar behavior on Windows devices and raised concerns about GDPR compliance and an estimated 30,000 tonnes of CO2e emissions if deployed to 500 million devices. Google later clarified that users can disable the model in Chrome settings.
Key takeaway
For AI Product Managers evaluating on-device model deployment, you must prioritize explicit user consent and transparent communication regarding file downloads and their purpose. Consider the environmental impact of large-scale model distribution, as a 4GB file across 500 million devices could generate 30,000 tonnes of CO2e. Ensure clear opt-out mechanisms are available and easily accessible to users to maintain trust and comply with privacy regulations like GDPR.
Key insights
Google Chrome allegedly downloads a 4GB AI model file without explicit user consent, raising privacy and environmental concerns.
Principles
- User consent is critical for data downloads.
- On-device AI models have environmental footprints.
In practice
- Disable Chrome's AI features via chrome://flags.
- Utilize enterprise policy tools to manage Chrome AI.
Topics
- Gemini Nano
- Google Chrome
- Data Privacy
- GDPR Compliance
- Environmental Impact
Best for: CTO, Executive, AI Product Manager, Tech Journalist, Policy Maker, AI Ethicist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Dataconomy.