Anthropic Says It Won't Stop Watching How Customers Use Claude. Check What Your AI Vendor Keeps.
Anthropic co-founder Benjamin Mann told Fortune on Wednesday, at the company's event in Singapore, that Anthropic won't hand over all oversight of how customers use its most powerful models. His reason is that the technology cuts both ways, and the company wants to catch misuse when it happens. Anthropic has also been publishing reports about people trying to use Claude for harmful things, including a group that allegedly tried to get help building ballistic missiles.
That's a company talking about governments and weapons, and I'm not going to pretend it's about your plumbing supply shop. But the same sentence has a smaller meaning for you. The AI company you pay every month is looking at how you use the thing, and it decides how long it keeps what you typed.
Every AI vendor does some of this
OpenAI, Google, and Microsoft all do some version of this. They watch for abuse, and to do that they store prompts and conversations for a while. Some keep them for weeks and some for a lot longer, depending on which plan you're on and which boxes got checked when someone signed up.
Anthropic did mention that it launched a program in September with stronger privacy for business customers, including zero data retention for eligible ones, while it keeps its misuse detection running. So the option to keep less exists. The part I see in the field is that nobody at a small company knows if they're on it, or if they're even eligible.
When I run a workshop here in Toledo, I ask the room a simple question. What's the most sensitive thing anyone on your team has pasted into an AI tool? Customer lists come up, and so do pricing sheets, a lawsuit letter, an employee review, and one time somebody's tax return. Most of the time the owner had no idea until I asked.
The fifteen minute check
First, write down every AI tool your team uses, including the free ones people signed up for with a personal email. Those are the ones you have the least say over. Then, for each paid tool, find the privacy or data controls page and look for three things: how long chats are kept, whether your conversations are used to train the model, and whether there's a business tier that turns that off.
If you're on a personal plan for work stuff, that's usually the quickest fix. Business plans often come with better defaults, and the price difference is small next to what's sitting in those chats. After that, give your staff one rule in plain words: no customer social security numbers, no bank details, no medical info, and no one's private records go into a tool unless you've checked the settings on it first.
You don't need a lawyer for this first pass. You need an hour, a list, and somebody willing to read a settings page. I'd rather you find out what your vendor keeps this week than when a customer asks you about it.
I run AI workshops and one-on-one AI consultations for companies around Toledo, Northwest Ohio, and Southeast Michigan, and a review of which AI tools your team uses and what they keep is a regular part of that. If you want a second set of eyes on your setup, email me and we'll find a time.
Email Jayson