Firms looking at AI research and drafting tools are asking the right first question: is this good enough to use. The harder one is what happens to the work after the tool produces it.
Harvey is an AI research and drafting assistant built on third-party language models, positioned for research, drafting, and document analysis. The vendor side of the contract covers data handling, model choices, and retention, and it deserves a close read before a pilot becomes a subscription.
None of that transfers the professional duty. Confidentiality runs from the client to the firm, and it does not move to a vendor because the vendor signed a data processing agreement.
If a client asks after the fact whether their matter was drafted with AI, or a carrier asks during a claim, the firm that answers well is the one that can produce three things: a written policy with an approval date, an approved tool list, and a review note on the matters where AI output was used. The firm that answers badly has a subscription receipt and a memory.
A tool inventory is not paperwork for its own sake. It is the difference between knowing which client data left the firm and guessing.
The free check includes a section on AI use and confidentiality. It asks whether there is a written policy, an approved tool list, and a review step, then scores the answer the same way it scores trust accounting and deadlines. It will not tell you which AI tool to buy. It tells you whether your firm could show the review if anyone asked.
The free check scores trust accounting, confidentiality and AI use, conflicts, and deadlines in about ten minutes. No client data, no account required to see the result.