Is our data safe with generative AI?
Yes, if the architecture and the contracts were chosen for that. The real risk often comes from unsupervised use, such as a confidential document pasted into a consumer tool. Supervising works better than banning.
Updated on 24 October 2026
What is going on.
The concern is legitimate: giving documents to a generative AI means sending them somewhere. Some companies have responded by banning consumer tools, others by signing enterprise contracts that are supposed to settle everything.
Both responses leave a blind spot. Business offers generally provide that your data is not used to train the models and specify where it is processed: you still need to check, and choose the architecture accordingly. But the main risk often lies elsewhere: with the person who, for lack of an approved tool or clear rules, pastes a contract or a customer file into a free service, from their phone.
What we observe.
- Banning pushes use towards uncontrolled tools.Teams who find a benefit in them carry on, but without saying so.
- Guarantees differ from one offer to another.The same provider can offer a consumer version and a business version with very different commitments.
- Rules of use are rarely written down.Everyone decides alone what they can give the tool.
What you can do right now.
- Provide an approved tool whose contractual terms you have checked.
- Write down on one page which data can be given to it, and which never.
- Ask teams which tools they really use, without looking for someone to blame.