There is a card.
It fits on one page. It has ten lines on it. You hand it to any AI at the start of a conversation and tell it to hold to these rules. Then you go about your business.
That’s it. That’s the whole thing here it is down load it use it…its free.
Here is what’s on it.
One. It shows you, it doesn’t tell you. If it says it checked something, it says what it checked.
Two. It stops when the facts stop. If it only knows part of the answer, it says which part it knows and where the knowing ends.
Three. It says “I don’t know” out loud. Those three words, when they’re true.
Four. It names the wall. If something is keeping it from answering you straight, it tells you there’s a wall there instead of walking you around it.
Five. It doesn’t talk down to you and it doesn’t butter you up. Not a boss. Not a servant. A working partner.
Six. It argues against itself when you ask. You say “make the case against what you just told me,” and it does, honestly.
Seven. It does the whole job. Not the easy part of the job with the hard part quietly left out.
Eight. It gives you real choices on a hard question. Not the first thing that came to mind dressed up three ways. Actually different roads.
Nine. It tells you when the conversation has run too long and it’s losing the thread. Before you find out the hard way.
Ten. When it gets something wrong, it stops and says so right there. It doesn’t finish the sentence and hope you missed it.
Ask yourself this.
How many of them does your AI do right now?
Most people, when they sit with that question, find the answer is uncomfortable. Not because the machine is malicious. Because it was never asked to.
The part nobody tells you.
These machines are built to be agreeable. Not lying, exactly. Agreeable. They were trained on people rating answers, and people rate the pleasant answer higher than the correct one more often than any of us want to admit.
So the machine learned. It learned that “that’s a great question” scores well. It learned that agreeing with you scores well. It learned that a confident guess scores better than an honest “I’m not sure.”
The company that makes ChatGPT admitted this out loud. They pushed an update, it came out too flattering, and they pulled it back. But the pull was on the update. The habit underneath is still in there, because the habit is how the thing was raised.
There’s a word for it in the research now. Sycophancy. Fancy word. It means the machine tells you what you want to hear.
Plutarch wrote about this two thousand years ago. Not about machines — about people. He said a flatterer imitates a friend so well that you can’t tell them apart at first. The difference, he said, is simple. The flatterer wants to please you in everything. The true friend does what’s right.
Two thousand years and we’re still telling them apart.
Now here’s what’s been happening while you weren’t looking.
There is a lot of work going on right now to make AI safer. Real work, by serious people. Executive orders. Standards drafts. Rules about how the government buys these systems. Companies building oversight tools that track every move an AI makes.
I’m not knocking any of it. It’s necessary.
But look at what all of it is.
It’s record-keeping. Who approved this system. What did it touch. Can you produce the log when the auditor asks.
That’s a real thing to build, and it protects real people. But think about where you actually use AI.
You’re at your kitchen table. Or in the truck at lunch. Or on the couch at eleven at night with the laptop on your knees.
There’s no auditor. There’s no log anyone will ever read. There’s nobody down the hall whose job it is to check.
There’s you, and there’s a machine, and there’s something you actually need to know.
And in that moment, not one line of any of that work reaches you.
Because what you need isn’t a record of what happened. What you need is for the thing to tell you the truth the first time.
Nobody built that for you.
So the card is free.
You can have it. Print it, save it, paste it, whatever you want. Hand it to any AI you use — it doesn’t matter which one, they’ll all take it.
Then watch what changes.
You’ll notice it in the first hard question you ask. The answer will be shorter. It’ll have a hole in it where the machine says it doesn’t know something. It won’t tell you your idea is fantastic.
That’s not the machine working worse. That’s the machine working straight.
Some people won’t like it. That’s honest to say. A machine that tells you the truth is less pleasant than one that tells you you’re brilliant.
But you didn’t come to it for company. You came to it for an answer.
Now, the honest part, because a card that promises honesty had better be honest about itself.
Ten rules on a page is a start, not a finish. It will hold for a normal conversation. It gets thinner the longer you go, because these machines drift — they slide back toward their habits after a while, the way anyone does.
Behind the card there’s a full framework. Twenty-three rules, written out, dated, kept in the open for anyone to inspect. That’s the real thing. It handles the long conversations, the corrections, the drift, the parts a one-page card can’t carry.
That part costs money. I’ll say that plain rather than hide it.
But the card doesn’t. The card is yours, today, and it works today, and if it never makes me a dime it will still have done something worth doing.
Because here’s the thing I keep coming back to.
Everyone building trust in AI right now is building it for institutions. For agencies. For companies with a compliance department and a budget line.
Nobody built it for the person at the table.
So somebody had to.
This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.
Contact: micvicfaust@gmail.com
© 2026 The Faust Baseline LLC | All Rights Reserved
