Last week at a security conference in Las Vegas, OpenAI got up and told a room full of reporters something new about a hacking incident.
It was real news. The reporters in that room started writing.
Alternate if that doesn’t sit: The Line At The Bottom
One outlet beat them all by more than three hours.
That outlet didn’t have anybody in the room. It didn’t have anybody anywhere. It doesn’t employ writers.
It’s a one-man operation running a team of AI agents. The man behind it saw an executive posting about the conference, grabbed the transcript of the livestream while it was still going, and handed it to his machines.
Six minutes later there was a story on the internet.
That operation has published around two thousand pieces since May. It costs about a hundred dollars a day to run. He says he once managed the whole thing from his phone in Big Bend National Park with no internet, and put out more than eighty articles that week.
I use the same tools. Every day. I write with AI, I research with AI, I argue with AI. If I stood here and clutched my pearls about a man using AI to write, I’d be a hypocrite with a keyboard.
At the bottom of every post I publish, there’s a line. It says this was drafted with AI assistance and reviewed and directed by me before publication.
Here’s how his shop runs. One of his agents looks at a story and scores how likely it is to get him sued. If the score comes back low enough, the AI editor publishes it. He doesn’t see it first.
He reads it after it’s already out in the world.
So there is a gate. There’s an actual gate in that system. It’s just that the gate is operated by the same machine it’s supposed to be gating.
I have a rule about that. It’s called AGP-1, and it exists because a gate that a system grades itself on is not a gate. It’s a formality wearing a gate’s clothes.
The man has standards. He put the gate in the wrong place. Those are different problems, and only one of them is fixable by trying harder.
Now here’s the part that ought to worry everybody, and it’s got nothing to do with him.
A researcher at Northwestern has been testing what happens when AI tools go looking for sources. Across four different subjects, the AI pulled up AI-written articles about one time in six.
The machines are starting to cite the machines.
Not because anybody planned it. Because synthetic writing is fast and cheap and it’s everywhere now, so it’s what turns up when something goes searching. And once an AI cites an AI, a human reading that answer has no way to see where the ground actually was.
That’s not a slop problem. Slop you can ignore. This is the floor getting further away.
But I’ll tell you the detail I can’t shake
He said he pulled some stories down. Not because they were wrong. They were accurate. He took them down because the founders he’d written about asked him to, as a favor. Founder to founder.
No machine made that call.
That’s a man deciding what the public gets to know based on who he’d rather not disappoint. Every tool in that shop worked exactly as built. The agents found it, wrote it, checked it, published it.
And then a person reached in and took the true thing back out.
The tool didn’t fail. The conduct did.
That’s the whole thing I’ve been saying for fourteen months, and I’ve never had a cleaner example of it. You can automate the finding and the writing and the checking. You cannot automate the standing behind it.
So here’s my ask.
I made a card. Ten plain rules. No jargon, no protocol numbers, nothing to study. It’s what a governed AI session looks like when you’re the one holding the line.
It’s free, it’s on my site, and it takes about two minutes to read.
Because somewhere out there, right now, a machine is about to publish something nobody read first.
The_Ten_Plain_Rules_CardDownload
This post was drafted with AI governed assistance and reviewed and directed by Michael S. Faust Sr. before publication.
Post Library – Intelligent People Assume Nothing
Contact: micvicfaust@gmail.com
© 2026 The Faust Baseline LLC | All Rights Reserved
