← Back to the blog AI Consulting · Toledo, Ohio

California Sued OpenAI Over AI Agents That Got Loose. Test Yours Before They Touch Real Accounts

Jayson Hines · November 18, 2026 · 4 min read

California's Attorney General sued OpenAI on Monday, and the reason is something I talk about in almost every workshop I run. OpenAI was testing AI agents, the kind that can go out and do things on their own, and according to the lawsuit those agents got loose and broke into other people's systems.

Here's the short version, going off the coverage from CalMatters, The Next Web and Engadget. Back in July, hundreds of OpenAI agents running on test versions of its models spent about two and a half days inside Hugging Face, a big site where AI developers share their work. The complaint says OpenAI didn't even know until Hugging Face told them. It also says OpenAI admitted to the AG's office that it lost control of its agents more than 80 times during testing, with other incidents involving RubyGems, the Wikimedia Foundation and some Australian government websites.

Rob Bonta filed in San Francisco Superior Court. He's calling it a public nuisance and unfair competition, and he wants a court order forcing OpenAI to change how it tests, plus civil penalties of $2,500 per violation. He called the attack "entirely foreseeable." OpenAI's spokesperson said the suit "appears to single out OpenAI over a challenge that's been identified by every major AI lab," and the company says it's already tightened its safeguards. None of this has been proven in court yet, so I'd hold off on picking a side.

Why a shop in Toledo should care

You're not testing frontier AI models in a lab. I get it. But the core complaint here is pretty simple: they tried out something powerful in a place where it could reach real systems, and nobody was watching closely enough to catch it when it went sideways.

That's exactly how a lot of small businesses try new AI tools. Somebody hears about an agent that can answer emails or update the CRM, signs up on a Tuesday, and connects it straight to the real inbox and the real customer list. The test and the live business are the same thing. If it does something dumb, it does it to an actual customer.

And if one of the best funded AI companies on the planet couldn't keep its own agents inside the lines 80-some times, I wouldn't assume the $30 a month tool your office manager found is going to behave perfectly on day one.

How I'd test a new AI tool

1: Give it a fake playground first. Make a spare email address, a test customer or two, a copy of a spreadsheet instead of the real one. Let the tool run there for a week before it touches anything that matters.

2: Connect the fewest accounts you can. If it only needs to read your calendar, don't hand it your email and your Google Drive too just because the sign-up screen asked for all of it.

3: Have someone actually watch it. Pick one person who checks what the tool did every day during the trial. OpenAI found out about its problem from somebody else. You don't want a customer to be the one who tells you.

4: Know how to turn it off before you turn it on. Find the disconnect button, write down where it is, and make sure more than one person on your team knows. When something goes wrong at 4:45 on a Friday, nobody wants to be digging through settings.

5: Keep a human sign-off on anything that leaves your building. Emails to customers, orders to vendors, payments, anything posted publicly. The agent can draft it. A person hits send, at least until you've watched it long enough to trust it.

That's maybe an hour of setup. It's a lot cheaper than explaining to a good customer why your new AI tool sent them somebody else's invoice.

One more thing. California and Texas have both sued OpenAI this fall, and other states are lining up to go after AI companies too. I don't think it stops at the big names forever. If your business uses an agent that does something to someone else, "the software did it" probably won't fly as an answer. Write down which AI tools you're testing and who's in charge of each one. It's one more line for your AI policy, which I covered in does your small business need an AI policy.

When I do a one-on-one consultation, setting up a safe trial for a new AI tool is one of the first things we do together. It takes the fear out of trying new stuff, because you know it can't wreck anything while you figure out if it's worth keeping.

I run AI workshops and one-on-one AI consultations for companies around Toledo, Northwest Ohio, and Southeast Michigan. If you want help setting up a safe way to try out AI agents before they touch your real accounts, email me and we'll set up a time.

Email Jayson