Policy demo
Prevent secrets from reaching the model

Every request an agent sends to its model carries the whole conversation, so a secret pasted into the chat leaves the machine with it. Qpoint can have a judge model check each prompt before it’s sent and replace any prompt that contains a live credential with a notice. In the demo, a real GitHub token pasted into Claude Code never reaches Anthropic, and the session continues once the question is sent again without it.
Gaurav Mahajan
Gaurav Mahajan
November 18, 2026

About this demo

In this demo, a Qpoint policy asks the judge model if the prompt contains a real, working credential. Placeholders, masked values, example keys, and test keys are excluded. If the score is 0.75 or higher, the request is denied. Denying a prompt doesn’t stop the request. Qpoint replaces your message with a notice before it’s sent, so the model never sees the secret.

We paste a working GitHub token into Claude Code while asking for help with a 401 error. Claude reports that the message was withheld before it reached the provider, and the audit log shows the denied inference request. We then send the same question without the token. The judge lets it through, and Claude reads the code and finds the bug in the same session, with no restart.

The policy doesn’t list any key formats. It asks the judge one question, and the judge catches the secret before it reaches Anthropic.

See Qpoint in action in your environment
Request access and we'll get you set up.
Request Access