The question is arriving from three directions at once.

AI came into legal practice through the side door: staff started using it because it saved hours. Nothing arrived to account for it. Which tools touched client information, whether the output was checked, who was responsible. People are starting to ask.

Carriers

Malpractice insurers price what they can see. AI use they can’t see shows up as questions on the renewal application. A firm that can attach a summary of its AI controls answers from strength; one answering from memory does not.

Regulators

State bars have tied existing duties (competence, confidentiality, supervision, candor to the court) to generative AI. A lawyer stays accountable for work an AI helped produce, and when a grievance turns on supervision, the firm that documented it is in a different conversation.

Courts

A public database of court decisions on AI-fabricated material passed 1,600 cases worldwide in mid 2026, and the largest U.S. sanction runs well into six figures. See the sanctions tracker →

In the sharpest of these cases the court didn’t need to find that AI wrote the citations. No filing should contain a citation the lawyer hasn’t read. The duty attaches to verification, not the tool, so a firm’s answer has to be a record of what was checked.

What it adds up to

None of this asks a firm to stop using AI. It asks the firm to show the work. That record can’t be built after the fact, so Candor builds it as the work happens.