The fastest way to lose attorney-client privilege in 2026 is not carelessness — it is pasting a client file into a tool whose terms let it retain and train on what you submit.
Privilege is about who can access the material
Privilege protects confidential communications made for legal advice. The doctrine assumes the material stays confidential. When a consumer AI's terms grant it the right to retain inputs, use them to improve models, or surface them to human reviewers, the confidentiality assumption can break — and a court may find privilege waived.
The questions to ask any AI vendor
- Are my documents and prompts used to train your models, or any third party's?
- How long are inputs retained, and can I delete them?
- Is my workspace isolated from other customers (no cross-tenant access)?
- Is there an audit trail of who accessed what?
If the answer to the first question is anything other than a flat "no," assume the material could leave the privileged circle.
Confidentiality isn't a feature you add later. It's the constraint you design around from the first upload.
What privilege-safe AI looks like
A tool built for privileged work never trains on your matter, isolates each firm's workspace, declines the data-retention terms that have put privilege at risk, and logs every action so you can show a partner or a bar exactly who touched what. A claim like § Pol 02 in a security policy should map to a real, openable commitment — not marketing.
Christie cites every answer to your own files.
Get early access →Privilege is recoverable as a habit: choose tools whose defaults keep client material inside the privileged circle, and document that they do.
