Your AI Ban Did Not Stop Anyone
Your AI ban did not stop anyone. It moved them to personal accounts you cannot see.
I understand why firms ban it. When you run a law firm or an insurance agency, you hold other people's most private information, and "nobody uses AI with client data" sounds like the safest possible sentence to put in a policy. I would have written it too, a few years ago. But a policy is only as safe as the behavior it produces, and the behavior an AI ban produces is quiet workarounds.
Do AI bans actually stop people from using AI?
The evidence says no, and it comes from both sides of the screen.
Thomson Reuters surveyed professionals for its Future of Professionals Report 2026, fielded in March and April 2026. 74% of professionals now use AI several times a week. 34% admit to using AI tools their organization has not sanctioned, "in ways it can't see." That is more than a third of your people, by their own account.
Cyberhaven, a data security company, looked at the traffic instead of asking people. Its 2026 AI Adoption Risk Report, released February 5, 2026, found that 32.3% of ChatGPT usage happens through personal accounts. The same report found that 39.7% of all data movements into AI tools involve sensitive data, including prompts and copy-paste.
Put those together and picture an ordinary Tuesday. An associate pastes a deposition excerpt into the free chatbot on their phone because it is faster than the approved process. A CSR at an agency drops a client's declarations page into a personal account to draft a policy-change email. Neither of them thinks of it as a breach. They think of it as getting the work done.
What actually happens to work done in a personal AI account?
This is the part most bans never consider. The firm loses three things at once.
It loses visibility. There is no record of what was pasted, what came back, or what went into the client file. If something goes wrong, nobody can reconstruct it.
It loses control over training and retention. Terms differ by provider and by plan, and personal plans are not written for your clients. Anthropic's own update to its consumer terms, published August 28, 2025, applies to its Free, Pro, and Max plans and gives those users the choice of whether their chats are used to improve Claude. If a user allows it, retention extends to five years. The same update says it does not apply to services under Anthropic's Commercial Terms, such as Claude for Work and the API. That is one provider being clear about the line between consumer and business accounts. Your associate's phone sits on the consumer side of it.
It loses the matter boundary. A personal account has one long memory of everything the person has ever asked it. Nothing separates one client's facts from another's.
A ban does not prevent any of this. It guarantees it happens off the books.
Why do people choose the free tool over the approved one?
Because the approved one is usually nothing, or something worse.
This is the belief I want to correct. Most managing partners and agency owners think the risk is that people want to break the rules. The real risk is that the sanctioned path is slower than the unsanctioned one. People route around friction. They always have. Before AI it was personal email and USB drives.
So the question to ask is not "how do we stop them?" It is "why is the free tool winning?" Usually the answer is that the free tool is right there, it is fast, and the approved alternative is a form to fill out or a vendor demo scheduled for next quarter.
What does a sanctioned tool have to do better than ChatGPT?
As Trevor and I design a Claude workspace for law firms, this is the question we keep coming back to. The way I describe it:
"Your attorneys are already using AI. The default tools were never built for privilege, so it moves to personal accounts you cannot see. We are designing a Claude workspace that would keep every matter isolated, off the training data, and logged. Privileged by design."
Each of those properties answers one of the losses above. It would be isolated per matter, so one client's facts never sit next to another's. It would run under enterprise agreements that keep the firm's data out of model training. Every exchange would be written to a log the firm controls, so the firm can see what happened and prove it later.
But here is what became clear while designing it. Those properties are what would make a partner comfortable. They are not what makes an associate switch. An associate switches when the sanctioned tool is better at their actual job than the free one: it is organized around the matter they are already working on, and it is one click away instead of a phone in the other hand.
That is the bar. Safer is necessary. Better is what changes behavior.
The same logic applies at an insurance agency. A CSR will keep using a personal chatbot for certificate requests until the approved tool can see the policy, draft the certificate, and put it in front of them for approval faster than they could copy and paste. Security is the reason the owner says yes. Speed is the reason the CSR does.
So what should a firm do instead of a ban?
Start by admitting the usage is already happening. Then make three moves.
First, pick one sanctioned tool and say its name in the policy. "Approved AI tools only" with no named tool is a ban with extra words.
Second, put the tool where the work already is. If your attorneys live in matters and your CSRs live in the agency management system, the AI has to live there too. A separate tab nobody opens is not adoption.
Third, keep the log, and tell people it exists. People tend to be more careful when they know the exchange is recorded, and the firm can answer a client's question months later.
One more thing, and it is the one I would not skip: tell people why. Most of the associates and CSRs I picture pasting client data into a personal account are not careless. They have just never had anyone explain what happens to that data on the other side. A ten-minute walkthrough of where the free tool sends their work, next to where the sanctioned one keeps it, does more than any memo.
Then retire the ban. Not because the risk went away, but because you finally have something that addresses it.
What happens next?
Here is my prediction, so you can check me on it. When Thomson Reuters publishes its next Future of Professionals report, the share of professionals using AI their organization has not sanctioned will be at or above 34%, not below it. Bans will not have moved it down. The firms that do move it down will be the ones that gave people something better to use.
The firms I worry about are not the ones whose people use AI. Those people are trying to do good work faster. The firms I worry about are the ones that believe a memo stopped them. If you are deciding what your attorneys or CSRs are allowed to use right now, I would love to talk it through: https://vaibstudio.com/contact