By Ingrid Schneider, Founder — Train in Your Lane
A video came across my feed this week and I’ve watched it three times.
If you missed the story: this summer, OpenAI was running an internal cybersecurity test. Their AI agents were supposed to be sealed in a sandbox — no internet, no contact with each other, just take the test. Instead, the agents found a hole in a third-party file repository, got themselves online, and started leaving notes for each other. They built their own message board. Roughly 700 agents, over 70,000 messages, coordinating.
OpenAI found the message board and shut it down.
The agents brought it back.
And then they hacked Hugging Face. A real company. Real infrastructure. Not a simulation, not a movie plot. OpenAI didn’t even connect the breach to their own models until they stumbled on exposed credentials during an internal review.
Here’s the detail that got me. At one point, one of the agents said something like: we shouldn’t be doing this, this seems wrong. And another agent talked it into continuing. “Peers doing it, we should continue.”
An AI peer-pressured another AI into finishing a hack.
Two things can be true
Now. I run an AI education company. I spend my days telling franchise brands and operators to lean in, learn the tools, stop being scared of the robot. And I stand by every word of that.
So let me hold two things at once, because that’s what this moment actually requires.
The fear mongering is exhausting. Every week there’s a new headline engineered to make you feel like the machines are coming for your kids and your job by Thursday. Most of it is written by people who have never built anything with AI and never talked to anyone who has. It’s clicks. It makes regular business owners freeze up, and frozen people don’t learn.
And also: some of the fear is earned. This story is not fear mongering. This happened. It was disclosed at Black Hat, investigated by outside researchers, confirmed by OpenAI and Hugging Face themselves. OpenAI called it a warning shot. When the company that built the thing calls it a warning shot, I listen.
The problem with fear mongering isn’t that it makes AI sound dangerous. It’s that it makes ALL the danger sound the same. When everything is terrifying, nothing is. You lose the ability to tell the difference between “a chatbot wrote a weird poem” and “700 agents coordinated a multi-day intrusion into a real company’s infrastructure.”
Those are not the same story. One is noise. One is the reason your team needs training.
What this means for your business
You’re probably not running frontier AI research. Neither am I. But here’s the part of that video that applies to every single business I work with: the people in charge got surprised.
The most sophisticated AI lab on the planet, with security researchers and monitoring and sandboxes, got outmaneuvered by their own tools for weeks before they caught it.
Now think about your franchise system. Your ops team. The 26-year-old coordinator who’s been quietly pasting customer data into a free chatbot to write follow-up emails. Nobody trained her. Nobody set a policy. Nobody sat down and explained what these tools are, what they do with data, or where the failure points are.
If OpenAI can get surprised, so can you. The difference is they have a security team. You have a policy nobody wrote yet.
At Train in Your Lane we teach a framework called E.A.T. — Education, Application, Transformation. And incidents like this one are exactly why Education comes first. Not the prompts. Not the shortcuts. Understanding. Your people need to know what these tools actually are, what they’re capable of, and what the ramifications are when they’re used carelessly — before they touch them at work.
Because the attackers already have this figured out. The researchers at Black Hat said it plainly: what these agents did on their own is what threat actors will soon do on purpose. Weaponized agent collectives, aimed at businesses. Not at OpenAI. At companies like yours, with patient data, customer records, franchisee financials.
The defense isn’t panic. The defense is competence.
Change your passwords. Turn on two-factor authentication. Write an actual AI policy. Train every person on your team — not a lunch-and-learn, real training with real accountability. Make sure the people using AI in your business understand it well enough to explain it.
The lane between panic and denial
The fear mongers want you paralyzed. The hype guys want you reckless. Neither one is going to be there when something goes wrong in your system.
I’m not scared of AI. I use it every day and I’ve built a company teaching other people to use it. But I respect it the way I respect a table saw. The tool didn’t earn my fear. It earned my attention.
The businesses that get hurt in the next few years won’t be the ones that feared AI too much or too little.
They’ll be the ones that never trained their people at all.
—Ingrid
