What happened
01
The Garante announced the decision on 20 December 2024.
02
It found no adequate legal basis for using personal data in training.
03
It found no age verification to keep under-13s away from unsuitable content.
04
It ordered a six-month public campaign on how data is used and how to object.
Why this is a design failure
When people are not told how their words are used, they cannot say no.
The fair alternative
Explain data use in plain words and make opting out easy to find.
Do
Ask clearly before using chats to train models.
Don’t
Don’t hide the training opt-out deep in settings.
Design guide for this case
If you're designing a general-purpose AI chat assistant
Interface
Before the first chat, show a clear choice screen asking whether chats may be used to train models, with equal-weight Yes and No buttons.
Words
In the content guide: "Explain in one plain sentence how chats are used and how to object. No legal jargon, no burying it in the terms."
Flow
Add an age check at sign-up that keeps under-13s out, and let users change their training choice from the chat screen in one step.
Metrics
Don't count opt-in rate as success. Track how many users can correctly say how their chats are used after onboarding.
Testing
Ask 8 new users to find and switch off model training. Time it and note anyone who gives up or can't find it.
Before launch
01
Do users choose about training before their chats are used?
02
Can the training opt-out be reached in one or two taps?
03
Does sign-up stop under-13s from getting in?