The real AI test for lawyers is governance, not adoption

What is the partner risk here: missed work, or a brief with fake citations? I would take the first all day. The second is how you get a judge, a client, and your own committee asking why anyone signed off on the file at all.

That is why the current AI story in law is not about enthusiasm. It is about control. Gartner says corporate legal departments are underestimating AI’s impact even as they face flat budgets and rising workloads, and it says legal teams already treat AI as a top priority. HaystackID says generative AI has moved from experimentation to standard use in eDiscovery, with hundreds of customers using it for review and privilege and more than 100 million documents analyzed. That is not a pilot. That is practice.

The question now is whether they can prove they controlled it.

The market has already moved

The old mistake was to treat AI as a side project. Latham & Watkins is training lawyers through an AI Academy. Microsoft’s Corporate, External, and Legal Affairs organization, about 2,000 lawyers and compliance professionals, is using Harvey AI for legal and compliance work. Descrybe has launched an Open Connector so firms and legal organizations can build and control their own AI research tools. Watts Law Firm says its new AI case-qualification platform still keeps every case subject to lawyer review.

Each of those moves points the same way. The firms and legal departments that are spending now are not buying magic. They are building rules, workflows, and review layers around machine output. Even the clearest cash signal in the material fits that pattern: Gartner says companies will double legal technology budgets by 2028. Money is following governance, not the other way around.

The courts are done with excuses

HaystackID also pointed to Schulte v. LinkedIn Corp., where the court allowed search-term culling before AI-assisted review. That matters because it shows courts are starting to fit these workflows into doctrine instead of pretending the old manual model still describes how discovery gets done.

The sanction record says the same thing louder. Coomer v. Lindell brought a monetary fine after nearly 30 defective citations and cases that did not exist. In the Innocent Chinweze matter, a New York City lawyer filed a brief riddled with fake cases after using Microsoft Copilot and was fined and referred for discipline. Mata v. Avianca ended with a $5,000 fine for six ChatGPT-fabricated cases. Whiting v. City of Athens, Tennessee led the Sixth Circuit to sanction two lawyers for appellate briefs with over two dozen fake or misrepresented citations and facts. Shahid v. Esaam ended with a struck divorce order and discipline for Diana Lynch. This is no longer one embarrassed lawyer at a time. It is a pattern.

Write the rule before you buy the tool

That pattern is what in-house counsel and litigators should be reading into the current wave of AI adoption. A recent report from the National Center for State Courts says courts are issuing new guidance and resources on AI, and a UC Davis law library guide says courts, judges, and state bar associations are actively developing ethics rules, opinions, local rules, and guidelines on AI use in legal practice. Another report says individual judges, especially in federal court, are maintaining standing orders on AI, some requiring certifications.

So the next internal question is not whether your shop uses AI. It already does, or soon will. The question is whether the firm or department can produce the written rule, the citation-verification process, and the workflow redesign that goes with it. If you cannot show that, you are not governing the tool. You are hoping the tool behaves. That is a bad risk model, and every partner knows it.