The question is arriving from three directions at once.
"How does your firm use AI, and can you show it?" A year ago that was academic. It is quickly becoming a question firms are expected to answer on the record.
AI moved into legal practice through the side door. No committee approved it; associates and paralegals simply started using it because it saved hours. What never arrived alongside it was any way to account for it: which tools touched client information, whether output was checked, who was responsible. That absence went unnoticed while no one was asking. People are starting to ask.
Carriers
Professional liability insurers price risk on what they can see. AI use they can't see is exactly the kind of uncertainty that shows up as questions on a renewal application, and, before long, as a factor in the premium. A firm that can attach a clean summary of its AI controls is answering from strength. A firm improvising an answer from memory is not.
Regulators
Bar authorities across states have issued guidance connecting existing duties (competence, confidentiality, supervision, candor to the court) to the use of generative AI. The through line is responsibility: a lawyer remains accountable for work an AI helped produce. Guidance becomes teeth the moment a grievance turns on whether that responsibility was exercised, and the firm that documented its supervision is in a different conversation than one that didn't.
Courts
The sharpest edge is already here, and it is no longer anecdotal. A public database maintained by legal researcher Damien Charlotin catalogues court decisions in which a party relied on AI fabricated material and a court responded; it passed 1,600 cases worldwide in mid 2026, having grown by roughly a hundred and forty in the preceding three weeks alone.
The penalties have moved as well. In April 2026 a federal magistrate judge in the District of Oregon sanctioned two lawyers a combined $110,204 over summary judgment briefs citing fifteen cases that did not exist, calculated at $500 per fabricated case and $1,000 per fabricated quotation, plus fees. In the Sixth Circuit, two attorneys were each fined $15,000 and held jointly responsible for the other side's appellate fees and double costs after filing briefs with more than two dozen fabricated or misrepresented citations.
That second case is worth reading carefully, because the court did not find the citations came from AI, and did not need to. Its point was narrower and harder to argue with: no filing should contain a citation the lawyer has not personally read and verified. The duty attaches to verification, not to the tool. Which is precisely why a firm's answer has to be a record of what was checked, rather than a policy saying AI is not used.
What it adds up to
None of these forces asks a firm to stop using AI. They ask it to be able to show the work, screened, supervised, and recorded. That record is dull to build after the fact and impossible to build retroactively. Candor's whole reason to exist is to make it accumulate quietly, in the background, so it's simply there when someone asks.