Scorecard — 10 honest questions about your AI

🪶 10 honest questions. Most AIs can't pass all of them — including ours, sometimes. The point is to know.

10 questions. One honest score. A signed gap report you can keep.

These are the questions a careful reviewer would ask about any AI you depend on — literacy, classification, oversight, incidents, evidence. Most AIs can't pass all of them, including ours sometimes. The point is to know.

MEOK AI Labs · Scorecard

10 questions. One honest score. A signed gap report you can keep.

Does everyone on your team who uses AI for work decisions know what it's good at, where it's wrong, and when to ignore it?

Have you named — out loud, in writing — which of your AI systems touch hiring, education, healthcare, identity, or critical infrastructure?

For each AI system that touches something that matters, do you keep a running list of what could go wrong — and re-check it whenever the system changes?

If the AI does something wrong, is there a real human in the loop who can stop it — and have you tested that person with a real failure, not a slide deck?

Before you let AI affect someone's life (a decision, a credit, a benefits cut, a job rejection) — have you written down how you'll know if it's biased, and what you'll do about it?

If someone from outside your company asked to audit your AI tomorrow, could you show them the receipts — not the slides, the receipts?

If your AI writes something — a letter, an image, a recommendation — does the person on the other end know it was AI, and can they verify that?

When your AI breaks in production — and it will — do you have a way to find out, fix it, and tell the people affected?

When your AI sees someone's personal data — health, finance, relationships — do you know exactly where it goes, who sees it, and whether the person agreed?

If you answered "yes" to anything above — could you prove it in a year? Could someone else verify the proof without trusting you?

Your score