TL;DR

A year ago legal teams mostly argued about which platform to buy and how many hours it would save. Now the questions are different: how the outputs get checked, how to keep sensitive data from leaking, how to tell it is actually helping, and above all who answers for it once the trial ends and people use it every day. Author Stephanie Corey writes that picking a vendor was the easier part. The rest comes after, and there is nothing new about it: running the change and setting the rules, not the software. And that, she says, is what legal operations already does, because it handles the same thing in every larger rollout where legal, IT, security, procurement and the business meet.

It is an opinion, though, not numbers, and the author runs a consultancy for exactly this work. The claim that legal ops is uniquely suited to the AI era is therefore also an ad for what she sells. Something still holds: whoever rolls out AI in a legal team runs into who watches it and who owns it sooner than into which tool to pick.

According to Harmonic Security, the most frequent AI users in companies are lawyers: 19.5% of all hours spent with AI, ahead of sales and marketing (17.7%) and developers (13.3%). The most used tool in legal teams is ChatGPT (67%), followed by Claude (18%), Copilot (9%) and Gemini (4%).

The numbers come from the AI Usage Index Report 2026 by Harmonic Security, a security firm headquartered in London and San Francisco, which measured 1,935,247 minutes of AI conversations across its enterprise client base over seven weeks (ending April 2026).

The tenth chamber of the Czech Supreme Administrative Court (NSS) received a cassation complaint that was, in the court’s words, largely text apparently generated by AI, and the attorney “did not take much trouble” revising it. The court went through the cited case law and found four kinds of defects: decisions irrelevant to the case, a correct case number with an invented date, a judgment attributed to the NSS that was in fact issued by the Regional Court in Brno, and one decision that does not exist at all.

The court labelled the whole thing “AI slop” and, unusually for judicial slang, added a footnote definition: low-quality content mass-generated by AI, often with no regard for factual accuracy.

No fine was imposed. The presiding judge told Info.cz the filing did not reach the intensity for one, but a stern warning was in order. The attorney admitted using a freely available AI tool: “It was probably some hallucination. I am taking a lesson from it.”

The decision’s load-bearing sentence aims at the whole profession: professionals trained in law take full responsibility for their filings, including any hallucinated AI argumentation. The Czech Bar Association is meanwhile talking to the leadership of the top courts and preparing its own guidance on AI use, expected in September. According to the Bar, courts intend to fine only hallucinating AI, not every filing they find to be poor work.

Decision 10 As 81/2026-27 sits, with verified citations, in the corpus of Czech AI case law (in Czech).