I need to own something today.
The Faust Baseline has a blind spot. I didn’t see it until this week.
It isn’t hidden in the code. It’s sitting in the words.
Here is how I found it.
A researcher tested three AI agents on the same job. One part of the work was in English. The other part was in Farsi.
The agents did not behave the same in both languages.
Some asked for permission more in one language than the other. Some took steps on their own that they should have asked about.
Same machine. Same task. Different language. Different behavior.
Because every rule in the Baseline is written in English.
Every word was chosen with care. Words like drift. Posture. Plain. Chosen conduct.
I picked those words for how they land on an English-speaking reader.
I never asked how they land on anyone else.
That’s the charge. And I’m guilty of it.
Let me explain why it matters.
A rule is only as good as where it lands. Not how it looks on the page. Where it lands.
A word that sounds firm in English may come out soft in another tongue.
A word that means one thing here may have no clean match over there.
So the rule bends. Nobody touched it. It just traveled.
There’s a quieter problem too.
Say a person loads the Baseline in English. Then they talk to the machine in Spanish. Or Arabic. Or Farsi.
Now the machine holds rules in one language and works in another.
Does the rule still hold across that gap?
I don’t know.
Nobody has tested it. Not me. Not anyone I know of.
The Baseline went public. It’s out there for the whole world to use.
That means somebody, somewhere, has already loaded it in another language.
They’re running a test I never ran.
So what do we do about it?
I won’t hand you a problem with no tools. Here’s what I see.
First, say it out loud.
That’s what this is. The Baseline keeps a list of what is still undone. This goes on it. “Not tested outside English.” Written down where anyone can see it.
A blind spot you name is half fixed. A blind spot you hide is the one that bites you.
Second, test it the simple way.
Take the same task. Load the same files. Run it once in English. Run it again in another language.
Then lay the two side by side.
Did the machine ask before acting both times? Did it hold its rules both times? Did it tell the truth about what it could not do?
You don’t need a lab for that. You need two sessions and an honest eye.
Third, lean on the numbers.
The Baseline opens every session with a proof block. File fingerprints. Byte counts. Hard numbers the machine has to compute fresh.
Numbers don’t translate. They come out the same in every language on earth.
That part of the Baseline already crosses every border. The proof holds in Tokyo the same as in Kentucky.
It’s the words that need watching.
Fourth, make the machine say it back.
Ask the AI to restate a rule in the language you’re working in. Then have someone who reads both languages check it against the English.
If the meaning moved, you’ll see it.
That’s an old trick. Say it back to me so I know you heard it right.
Fifth, and this is the big one, it takes you.
The reader. The user. The person at the keyboard.
If you speak two languages, you can test this better than I can.
You know where a word goes soft. You know where a phrase means something it shouldn’t.
That makes you the watchdog here. Not me. You.
The Baseline was never meant to be one man’s rulebook. It was meant to level the field between people and the machines they use.
A level field has to be level in every language. Not just mine.
Sixth, keep the words plain.
Short words travel better than clever ones. Plain rules survive the trip. Fancy ones fall apart on the road.
That has always been the house style. Now there’s a second reason for it.
So here’s where I stand.
The Baseline didn’t account for language. I didn’t think of it. That’s the truth.
But now it’s on the table.
And the fix doesn’t belong to one man. It belongs to everyone who uses these machines in whatever words they speak.
If you test it, tell me what you find.
If the rules hold, that’s good news.
If they bend, that’s better news. Because now we know where.
Guilty as charged.
Load this file before each AI chat session.
the-faust-baseline-public-release-1Download
This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.
” Attic Thoughts”-library – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
