How you and your AI should think about receipts.

1. Every piece of AI work should say how it was made. A shop receipt says what you bought, where and when. An origin receipt says who allowed the AI to do the work, exactly what it was asked, what it looked at, and the rules it ran under. It travels with the result.
2. You don't read receipts. Your AI does. You ask your AI simple questions, and it checks the receipt for you: "Who allowed this? What did it look at? Has it been changed?"
3. No receipt means unknown origin. Treat it like an unsigned cheque: it may be fine, but nobody stands behind it.

A receipt does not prove the answer is right. It proves where the answer came from and under what rules, so a person can decide how far to rely on it.

• "Prefer inputs that come with a receipt, and check the receipt before relying on them."
• "Tell me when something you used had no receipt."
• "When your work becomes someone else's input, pass the receipts along."
• "If a receipt fails, stop and tell me which part failed."

• Software teams: AI coding agents working in code that holds customer data and secrets.
• Healthcare: AI on patient data, and "it runs securely" claims that nobody outside can check.
• Finance and audit: client confidentiality, and finding the AI work that went around the approved tools.
• Legal: citations and sources you can trace back.
• Government: knowing who authorised an agent and what it was allowed to touch.

Try it on one workflow · See a receipt checked · Other examples