AI Protect
Shows a firm which AI tools its team uses, and catches client data at the paste, before it reaches any of them.
- Role
- Led it end to end
- Team
- Four people
- Built at
- NEXTech, 2026
- Runs on
- macOS and Windows
The problem
People paste client work into AI tools, and most of those tools were never approved by anyone. The product site cites a Teramind report: more than eight in ten employees use AI tools nobody signed off on.
For a CPA firm, a law office or a dental practice, that is a tax return or a patient record in a chat window. These firms rarely have a security team, and most owners can't say which AI tools their people use. Banning AI doesn't work either: the tools help, so people find a way.
Caught at the paste
AI Protect answers two questions. First, who is using what, and whether you approved it. Second, what just got pasted or uploaded into one of those tools.
The firm decides what counts as sensitive: client records, financial details, medical information, or its own lists. When a paste matches, the risky part is held back and everything else goes through, so the work still gets done. Try it below.
An AI assistant, any of them
Whitfield_1040_2025.pdf
Draft a summary email for this client. Their SSN is 078-05-1120 and the refund goes to account 4417 2290 0183.
AI Protect
Press Send. The check runs on this computer before anything leaves it.
Private by default
A tool that watches what people paste could easily become surveillance. We drew the line in the product itself. The check runs on the device, and the console learns what happened, not what was said.
Never leaves the computer
- The text or file being pasted or uploaded
- Keystrokes, screens, browsing history
- Anything typed that isn't a paste into an AI tool
What the console receives
- Which kind of sensitive data was found
- Who, which AI tool, and what AI Protect did
- A masked snippet of the surrounding text, only if you turn it on
- Which assistant and model a person used, only for people you opt in
The decisions that mattered
Coach before you block
Every rule ends in one of five actions: allow, log only, coach, redact or block. Coaching shows people why something was held and what to do instead, like pointing them to the firm's approved AI workspace.
Policy follows the person
A rule is data type × destination × channel, and the first match wins. Policy is assigned to a person, not a machine, so it follows them across devices and reaches each one within a check-in.
No green it hasn't earned
The console shows every device's real state, down to the agent version. A device that has gone quiet shows Unknown, never a reassuring green.
MDM is optional
Small firms often have no device management. One emailed link per employee installs it; firms that do have Jamf, Intune, Kandji or Hexnode push the same signed installer.
My part
I led AI Protect end to end with a team of four: the product thinking, the decisions, and the build across the endpoint agent, the browser extension and the console.
It ran on the same workflow as my own projects. Decisions were written down before code, tests came first, and agents wrote a lot of the code under review. The product is commercial, so this page sticks to what its public site already says.