What happened
01
The Meta AI app launched in April 2025 with a public Discover feed.
02
Reporters found medical, legal and personal questions posted publicly.
03
Mozilla and privacy researchers called for the feed to be changed.
04
Meta added a warning around 16 June 2025: prompts you post are public.
Why this is a design failure
When sharing looks like saving, people publish what they meant to keep private.
The fair alternative
Keep chats private by default, and make publishing a separate, clearly labelled step.
Do
Show exactly who will see a post before it is shared.
Don’t
Don’t place “share” where people expect “save”.
Design guide for this case
If you're designing an AI chat app with sharing
Interface
Keep "Save" and "Share" as separate buttons in different places, and label the share button with who will see the post.
Words
In the share screen copy: "Say exactly who will see this, for example 'Anyone on the app can see this post.' Never just say 'Post'."
Flow
Keep chats private by default, and make publishing a separate step with a preview and a confirm that names the audience.
Metrics
Don't count posts to the public feed as success. Track how often people delete or make private a post soon after sharing it.
Testing
Ask 5 users to save a chat for later, then ask who can see it. Every answer should be "only me".
Before launch
01
Are chats private by default?
02
Does the share step show exactly who will see the post?
03
Is "share" kept away from where people expect "save"?