Rule 1.6 when the tool is a third party

1 · Learn the move · Constraints + negative instructions

Rule 1.6 does not ask whether the tool is good. It asks whether you revealed information relating to the representation, and to whom [M]. A consumer chat window is a third party: the vendor can see what you typed, may keep it, and under some plans may use it to train [S]. The bar opinions that have addressed generative AI say roughly the same thing in different words: the duty of competence covers understanding the tool, the duty of confidentiality covers what you put in it, and for a self-learning tool a general engagement-letter consent is not enough [P]. Privilege is a separate question with the same shape: a communication shared with a third party outside the privilege can lose it, and whether a vendor under contract is inside or outside is argued, not settled [S]. So the constraints go into the prompt, not into your memory of them. The model is told what never comes back (a name, a matter fact that identifies, a privileged communication) and what it does when the paste carries one (stops, says so, returns nothing else). Negative instructions are the whole lesson: the list of what the output must not contain is longer than the list of what it must.

Before I paste anything about a client, answer three questions from the plan terms I paste below: (1) who at the vendor can see what I send, (2) whether it is used to train a model, and (3) how long it is kept. Quote the line that answers each or write [NOT IN THE TERMS]. Do not tell me whether this plan is safe for client information, whether a paste would be a Rule 1.6 disclosure, or whether privilege survives; those are my questions and my bar's. Terms: [paste]