AI Policy for Good Attorneys At Law, PA
(1) Privacy, security and confidentiality come first: I am fully aware that open AI models train those models on the data you feed it. I assume so even if they say they don’t. I know that if the price is free (or even sometimes if you pay), you’re the product. As such, I never enter into any open AI model information or documents that contain a client’s name or other identifying information, or sensitive information about their business or other affairs. My prompts are “generalized”, so that they can’t be associated with anyone in particular, or research-oriented, which just asks for law and analysis that’s publicly available.
(2) I do not use AI content as final work product: AI is better each day at producing draft agreements and so forth, but I use the output only as a model, template, or basis for further development. My clients never receive canned content that I have not personally reviewed and edited.
(3) Facts, cases and statutes are always verified: Hallucinations and errors, while WAY less frequent than they used to be, are still a thing. Fortunately, the models have gotten a lot better at footnoting the text with the original sources of the information. I always review any case law, statutes, or articles to see that what was produced in the AI summary is legitimate and correct.
