Well, defensibility in litigation means something more specific than it does in most other professional contexts.  As you rightfully mention, there is hyper focus on it: every assertion made must be capable of withstanding challenge from opposing counsel and the court. When AI is part of the process that produced those assertions, the challenge extends to the methodology itself, not just the conclusion.

In practice, defensible AI in litigation requires three things. First, source traceability: every statement made must point back to a specific passage in a specific document that a human can locate and verify. Second, process reproducibility: the same methodology must have been applied consistently across the entire document set, in a way that can be explained and demonstrated if the review is challenged. Third, human oversight accountability: a qualified attorney must have made the ultimate judgement on outputs that matter.

Many lawyers still think of defensibility in the context of AI as a disclosure question. They only care about whether they need to tell opposing counsel or the court that AI was used. In my opinion this is the wrong frame. Defensibility is about methodology, not disclosure. The practical implication is that lawyers need to be able to answer: what system was used, what it was asked to do, how its outputs were reviewed, and by whom. Well-built legal specific technology tools should accommodate for this. 

But enthusiasm has to be grounded in ROI, and when you do the honest accounting on drafting efficiency, the numbers consistently favour starting from a trusted, expertly-crafted precedent paired with a deterministic approach to drafting. The comparison that matters isn’t “AI output vs. a blank page” – it’s “AI output including all downstream review, supervision, and error-correction time” vs. “a logic-driven automation built on solid precedent.” When you run that calculation, the deterministic, precedent-first approach wins. Every time a lawyer has to interrogate an AI-generated clause they didn’t author, you’re paying a tax. That tax compounds across a large practice.
So the question we’d put to firm leaders isn’t “are you pro-AI or anti-AI?” It’s “where in the workflow does AI actually move the needle on ROI, and where does it just move the risk around?”