It happens more than anyone likes to admit. A contract lands on your desk, 30 pages of dense language, and you're short on time. Pasting it into ChatGPT to summarize or explain a clause feels like a harmless shortcut.
It usually isn't.
What actually happens to that document
When you paste text into a general-purpose AI chatbot, you're sending it to a system that, depending on the plan and settings, may retain that input, may use it to improve the underlying model, and almost certainly wasn't built with attorney-client confidentiality or client consent in mind.
Most consumer AI tools are designed for general use, not for handling material you have a professional or contractual duty to protect. This isn't a hypothetical risk. It's the default behaviour of tools built for broad, general use and not for confidential documents specifically.
Your client didn't consent to their contract being read by a third-party AI vendor. Did you check whether they needed to?
Why "it's probably fine" isn't a real answer
A few honest questions worth sitting with:
- Did your client consent to their contract being processed by a third-party AI vendor?
- Do you actually know whether that vendor's terms allow your input to train their models?
- If asked in an audit or a malpractice inquiry, could you explain exactly where that document went and what happened to it?
If any of those give you pause, that's the signal. The convenience of a quick paste-and-summarize isn't worth the exposure if you can't answer those questions clearly.
What "safe" actually requires
For a tool to be genuinely safe for confidential documents, it needs to satisfy a few concrete conditions and not just claim to be "private" in its marketing:
- 01 A clear, contractual commitment that your data is never used to train models. Not a vague privacy policy but an actual term you can point to.
- 02 No retention beyond what you control. You should be able to delete a document and have it genuinely gone, not just hidden from view.
- 03 Verifiable answers. If an AI tool summarizes a clause, you should be able to check that summary against the original text in one click rather than just taking its word for it.
- 04 Built for this specific use, not repurposed from a general chatbot. A tool designed around document confidentiality behaves differently, by default, than a general assistant with a privacy policy bolted on.
Where Private fits
Private was built specifically around these conditions. Documents are never used to train any model, and that's a contractual guarantee rather than a marketing claim. Every answer cites the exact page it came from, so you're never just trusting the AI. And deleting a document removes it immediately, not eventually.
It's not a general chatbot you're hoping behaves responsibly with sensitive material. It's built for exactly this kind of work: asking questions of documents you have a duty to protect.