Perspectives
Banning ChatGPT does not keep your data in
People paste where the work gets done fastest.
By Ehsan Gazar, 23 October 2026
Somebody in finance has a contract to summarise by three o'clock. They paste it into ChatGPT on their own phone, because it is quicker than anything the company has given them.
Nobody sees it happen. The company has no record of it, no agreement covering it and no way to ask for it back.
So the usual answer arrives: a memo, and a blocked website. A month later the same contract goes into the same app, on a personal laptop at home.
The ban did not keep the data in. It only moved the paste somewhere you cannot see.
Why the ban does not hold
People do not paste confidential files into a chatbot out of carelessness. They do it because the chatbot helps, and the deadline is real.
A ban removes the help and keeps the deadline. The work still has to be done, so people find the nearest tool that does it.
That tool is a personal account. It sits outside your sign-in, your contracts and your records. Blocking it at the office firewall changes which network the paste travels over, nothing more.
There is a cost on the other side too. The people who obey the ban fall behind the people who do not, and both groups notice.
The question is not how to stop people using AI. It is how to make the safe way the quickest way.
If the company tool is slower or weaker than the free one on a phone, people will keep using the phone. An approved tool wins only by being at least as good at the task.
The Vatan Assistant is the workspace's own AI tool. People sign in with the identity provider the company already runs, such as Entra, Okta or Google Workspace, on a domain the company has proved it owns.
It offers models from several providers rather than one, so nobody needs a second app because a different model writes better. It can also answer from the company's own documents, which a personal account never sees.
Who may use it is a permission, set per team, and each team can be given a budget of its own.
That is the part a memo cannot do. It gives people a better place to work, with the rules built in.
Guardrails on the way out
Rules about what may be pasted work best where the data leaves the company, not only in a document people read once.
Every chat message the Assistant sends to a model passes through the Vatan gateway, the same one the company's own applications use. Redaction is set there, once, for the whole workspace. It applies to everybody in it, without anybody configuring it person by person.
Redaction looks for identifiers with a fixed, checkable shape. It finds email addresses, payment card numbers that pass the Luhn check, IBANs and UK National Insurance numbers. It also finds US Social Security numbers and common API key formats.
In redact mode, each one is replaced with a typed placeholder such as [EMAIL] before the message leaves. The model can still work with the sentence, and the address never reaches the provider.
In block mode, a message carrying one is refused whole, and nothing is sent to a provider at all. The refusal names which kind of identifier stopped it, never the value itself.
Both modes are off until somebody switches them on. A name or a diagnosis has no fixed shape, so those stay the job of what people are taught.
Where the data goes, and what is kept
A prompt goes to the company that serves the model the person chose, and every such company is named on the sub-processor list. Vatan trains no models on it.
Conversations a workspace chooses to keep are encrypted with a key that belongs to that workspace alone. That key is itself encrypted with another that never sits in the database.
The usage record for each request holds the tokens, the cost and the timing. It never holds the prompt or the answer. The company chooses how long those records are kept.
So a manager can see which teams use AI and what it costs, without anybody reading what was asked. The full account is on how we handle your data.
Where a team starts
The useful first step is small. Pick one team that already pastes into a personal chatbot, and ask them which three tasks they use it for.
Give that team the Assistant with redaction switched on, and set a monthly budget. Then let them do those three tasks there for a fortnight.
At the end, ask what was missing. If they still reached for their phone, that names the gap. It might be a model, a document or a step that takes too long.
Close that gap and the ban becomes unnecessary, because the company tool is already the quickest way to the answer.
Keep the memo, but make it one line. Use the company tool for work, and here is how to sign in.
People keep data in when the safe tool is the one that gets the work done.