Google Chrome Installs 4 GB Gemini Nano AI Model Silently – No User Consent

Google Chrome is silently installing its Gemini Nano AI model — weighing in at roughly 4 GB — on user devices without asking for permission. The model, part of Google's on-device AI push, is being downloaded in the background and stored locally, which can eat up significant disk space on unsuspecting machines.
How It Works
According to reports, Chrome is downloading the model as part of its built-in AI features (like smart compose or summarization), but the installation happens automatically. Users are not prompted, and there is no clear opt-in flow. The model is stored under Chrome's local data directory, typically ~/.config/google-chrome/GeminiNano/ on Linux or the equivalent on other OSes.
- Size: ~4 GB for the full model download.
- No Consent: The download begins without user interaction or notification.
- Background Process: Happens via a Chrome updater service or built-in component updater.
What You Can Do
If you want to remove the model or prevent the download, you can:
- Disable Chrome's AI features via
chrome://settings/safetyCheckorchrome://flags/#optimization-guide-on-device-model. - Delete the model folder manually:
rm -rf ~/.config/google-chrome/GeminiNano/(Linux/Mac) or the corresponding Windows path. - Use an enterprise policy: Set
OptimizationGuideOnDeviceModelEnabledto false.
This is a significant privacy and storage concern, especially for users on limited data plans or constrained disk space. Developers and power users should check their systems immediately.
📖 Read the full source: HN AI Agents
👀 See Also

When Online Commenters 'Detect' My Art as AI: David Revoy's Frustration
Artist David Revoy shares screenshots of online commenters falsely accusing his hand-drawn art of being AI-generated, despite sharing timelapses. He explores the problem in his comic 'Authenticity Problem'.

Claude Code Opus Fails with Rate Limit Error Despite Available Weekly Capacity
A Claude Max subscriber reports that Claude Code Opus returns 'API Error: Rate limit reached' even though their usage dashboard shows 97% of their weekly 'All models' capacity remains unused. The issue occurs specifically in Claude Code while Opus works normally on claude.ai from the same account.

Claude Code v2.1.85 Release: MCP Improvements, Hook Filters, and Bug Fixes
Claude Code v2.1.85 adds environment variables for MCP headersHelper scripts, conditional if fields for hooks to reduce process spawning, and fixes for /compact failures, plugin enable/disable issues, and terminal keyboard problems in Ghostty, Kitty, and WezTerm.

Elevated Errors on Claude Opus 4.7: Status Update and What to Expect
Claude Opus 4.7 is experiencing elevated errors as of 2026-05-19T15:21Z. Check status.claude.com for progress and resolutions.