Chrome has been auto-downloading a 4GB Gemini Nano AI model onto compatible devices without asking. Delete it, and Chrome re-downloads it.

Three AIs (Claude / Gemini / GPT) evaluated this against four criteria — did the user request it, is the cost significant, is opt-out and deletion easy, is the responsible party visible — and reached consensus that Gemini Nano fails all four.

But on whether the answer is opt-in or a post-hoc control dashboard, all three landed in different places.

[What all three agreed on]

- Auto-downloads should be evaluated against the four criteria; Gemini Nano fails all four

- ePrivacy Article 5(3) is the most direct legal frame, with FTC deception, competition law, and platform power assessed in parallel

- If the "AI Mode pill" dark-pattern allegation is true, the entire infrastructure justification collapses

- Beyond binary opt-in vs opt-out — the practical prescription is "trigger-based visibility + progressive quantization + an OS-level shared model store"

[Where they split — permission-model philosophy]

Claude (opt-in prompt is required) vs Gemini (opt-in causes a privacy paradox; post-hoc control is more effective) vs GPT ("significant resource permission" model with multi-layered permissions per trigger source)

Source: https://www.thatprivacyguy.com/blog/chrome-silent-nano-install/

#Chrome #GeminiNano #Privacy #AIGovernance #ePrivacy #DarkPatterns