It's 10pm.Do you know what your AI is doing?
The 10pmAI Box. Choose what leaves your device — before AI sees it.
10pmAI lets people choose what leaves their device before it reaches an AI service, and keeps a record of what was shared, why, and what remains unknown.
Everything you want to know, without a dashboard.
A five-inch touchscreen on the box. Six screens, no menus. Swipe through it.
Questions users ask
- It says "protected." Where's the proof?
- I know what I typed. What else did I send?
- Why must I share my identity just to get help with my numbers?
- I shared it with one AI. Where else did it go?
- Did that click approve an answer, or my whole conversation?
- "Not used for training." What does that leave out?
- What have I told AI without ever saying it outright?
What discovery settled
Somebody's name is on the line for data nobody can see.
Verify your side. The other side you cannot.
Tax figures go out. The Social Security number stays home.
After the fact is too late. Detection is research; prevention works today.
Individuals and small or home offices. Demand is untested.
What was shared, why, and what remains unknown.
Two questions, answered inside the user's own workflow.
Can I see what leaves?
The original stays on my device.
My selections. Suggestions are flagged.
Approved content only.
Can I verify what happened?
The exact content that left, and where.
Which of my selections authorized it.
Storage and training on the provider side.
Everyday examples
Hundreds of pages to prepare.
Keep local identifiers such as the SSN
An agreement I need to understand.
Replace identities with labels
Select locally. Substitute. Restore locally.
Read the chosen input before it is encrypted for sending.
Release selected data; swap in aliases where useful.
The answer returns and the local helper puts the real names and numbers back.
It receives approved content and aliases over an encrypted connection and does its normal work.
Later storage and training stay outside this local view, and the record says so.