The fastest way to create an ethics problem with AI is also the most common: pasting privileged client material into a public, general-purpose chatbot. Before it is a question of accuracy, it is a question of confidentiality — where did that data go, and who can see it?
The mistake, and the fix
General consumer AI tools are not built for privileged material. The fix is not to avoid AI; it is to use tools designed for the job — ones that handle your files as the sensitive client data they are, read only what you choose to upload, and keep that evidence under your control.
Use tools built for privileged material, not a public chatbot.
What “built for it” looks like
A tool built for confidential case work reads only the file you upload — no browsing, no outside data — and lets you export and delete a matter when it closes. It is a working environment for active cases, not a permanent archive. And because it grounds every answer in your own evidence and links it to the source, there is nothing pulled in from outside and nothing to leak from beyond your file.
Practical steps
Keep confidential data out of general chatbots. Prefer tools that show their sources so you can verify. Ask any vendor how your data is handled before you upload. Our own approach is on the Security page.