This lease between Daniel ReyesTENANT and Harbor Point LLCLANDLORD for the unit at 212 Wexford AveADDRESS begins on…
Asking an AI to summarize a lease, find the termination clause or list every deadline in a settlement is one of the most useful things it does. It’s also where a lot of people quietly paste in a client’s name, address, account numbers and medical history.
That might be fine. It might also be a problem for your duty of confidentiality, your firm’s policy or HIPAA. The good news is that the AI almost never needs the identities to do the job.
What actually leaves your computer
When you paste text or upload a file to a chat assistant, the whole thing is sent to the provider’s servers. What happens next depends on your plan and settings: whether conversations can be used to improve the model, how long they’re kept, and who can review them. Business and enterprise plans usually have stricter terms than personal ones.
Check your own plan’s data controls before you rely on them. And remember that “not used for training” isn’t the same as “never stored.”
For lawyers, the ABA’s Formal Opinion 512 (2024) says you need to understand these risks before entering client information into a generative AI tool, and in some cases get the client’s informed consent. For healthcare, patient information generally shouldn’t go to a service that hasn’t signed a business associate agreement with you.
A five-minute routine that works with any AI
Ask what the question needs. "When can the tenant terminate?" needs the clause, not the tenant's name. Paste only the relevant sections when you can.
Replace identities with roles. Use Tenant, Landlord, Patient, Employee A. Be consistent: if "Reyes" becomes Tenant in one paragraph and Party B in the next, the answer will get confused.
Hunt the less obvious details. Account and case numbers, dates of birth, email signatures, headers and footers, and the file name itself. These are the ones people miss.
Check your settings. Turn off model training if your plan allows it, and prefer a temporary or business workspace for client work.
Keep the key on your side. Write down who "Tenant" is on your own computer, so you can map the answer back to the real people.
Or let the computer do it
Doing this by hand works, but it’s tedious, and tedium is where mistakes come from. We built Veil partly for this. It finds names, addresses and identifiers on your computer, swaps each one for a consistent placeholder like [PERSON_1], and only then sends the text to your own ChatGPT or Claude account. Answers come back with the placeholders, and Veil keeps the key.
What you type into forms and your saved profiles is never sent. A “What the AI sees” view shows the exact text before it leaves.
Whatever you use, the principle is the same: give the AI the problem, not the people.
This post is general information, not legal advice. Check your jurisdiction's rules and your own AI provider's current terms.