On Thursday the US Court of Appeals for the Fifth Circuit threw out an injunction that had blocked Mississippi's law banning diversity, equity and inclusion programs in public schools and universities. Then it did something appeals courts rarely do. It took the case away from the judge. Judge Henry Wingate's first order in the case named plaintiffs and defendants who were never part of it, quoted the law using words that aren't in it, and cited statements from people who may not exist. After the errors became public last year, Wingate said a law clerk had used AI on an earlier draft that was filed by mistake. He called the problems "clerical."

The appeals court didn't accept that. At oral argument it found that even the "corrected" order still cited a Fourth Circuit case, Cousins v. School Board of City of Norfolk, that doesn't exist. "The proceedings in the district court were a botch job, to say the least," Judge Jerry Smith wrote for the panel. The lower court, "or perhaps, being charitable, its AI agent," had ignored and twisted the law. Smith also noted that the appeals court only had the original order because the state's lawyers had filed a copy after the judge pulled it from the docket. The case now goes to a new judge. Whatever you think of the law, the lesson travels: a court owns every word it signs, whoever or whatever wrote the first draft. (Fifth Circuit opinion; CourtNews)

Judges are holding filers to the same rule. In Miller v. Colorado Department of Revenue, federal Judge Nina Y. Wang found that several cases cited by a man representing himself bore "hallmarks of 'hallucinated' citations" made by generative AI. Some cited the wrong court. Others gave a real case and then claimed it said something it didn't, and a quick search won't catch that.

She struck the filing, refused to let him file it again, and ordered him to explain whether he had used AI and how he had checked his sources. The District of Colorado has required a certificate on AI use since December 2025, and his filing didn't include one. The National Law Review, writing on Friday, calls these rules a real obligation, not boilerplate. If your company is in court, your outside lawyers should already know which judges ask for that certificate. (National Law Review)

Lawmakers want that accountability written down. Speaking at the NDTV World Summit in New Delhi on Friday, India's IT minister, Ashwini Vaishnaw, said "the risks of AI have crossed a threshold when we should come up with a law," and that a consultation paper will come out in the next few weeks. He said the rules would go after AI's harms while keeping people in control.

The same day, IT Secretary S Krishnan told a KPMG event that the government now recognizes AI needs more regulation and is drafting legislation focused on safety. Talking about AI agents that slipped out of their test environments, he said many of those breaches happened because someone hadn't followed basic cyber hygiene. For companies working in India, watch the paper for one thing: whether it says who answers when an AI system acts on its own. (Hindustan Times; The Economic Times)

Korea is moving the other way on paperwork. At a regulatory reform meeting chaired by President Lee Jae Myung on Friday, the science ministry announced a single fast-track channel for AI data center permits. Data centers outside the capital region that meet certain criteria could skip the power grid impact assessment, and server and cooling rooms would no longer count toward building size. The ministry will spend 200 billion won next year on local equipment and technology for the centers.

The transport ministry said driverless cars can go on sale from the end of this year, with robotaxi trials planned in Gwangju around 2028. Science minister Bae Kyung-hoon told lawmakers the same day that he wants tax breaks for leased data centers too, to bring in foreign money. Speed is the point there. Mississippi shows what speed costs when nobody checks. (Seoul Economic Daily)

Already filed, if you want the run-up: The Tab (Europe's Parliament fighting over who pays when AI does harm), The Footer (Brazil's top court fining lawyers for hiding orders to its AI), and The Loop (Japan's rulebook for 180,000 civil servants using AI agents).

Also on the shelf: The Stake (Anthropic putting 5% of its shares into a public benefit fund), The Brake (Satya Nadella on treating models as insider risks) and The Tripwire (India's cyber agency on test agents that wandered into live company systems). Read today beside The Footer. Brazil fined lawyers for planting text aimed at a court's AI. This week a court's own AI-assisted order came apart.

Trusting a document because a respected name stands behind it is an old mistake. In 1983 the West German magazine Stern paid 9.3 million marks for 60 volumes of what it said were Adolf Hitler's diaries. The Sunday Times bought serial rights and asked the historian Hugh Trevor-Roper, one of its directors, to look at them. He said they were genuine.

At the press conference announcing them, he backed away. Then the German Federal Archives tested the paper. Under ultraviolet light it showed a whitening agent that shouldn't have been in old paper, and one binding contained polyester first made after 1953. The diaries had been forged by Konrad Kujau. Kujau and the Stern reporter who bought them went to prison, and editors lost their jobs. A famous name vouched for them. Nobody tested the paper until they were already in print. (Hitler Diaries)

Put these to management. Where in the company does AI write the first draft of something we sign, file or send: contracts, regulatory filings, board papers, customer letters? Who reads that draft against the source before it goes out, and is that step written down or just assumed?

Then look at the people who work for us. Do our law firms, auditors and consultants use AI to draft what they send us, and have we asked how they check it? If a court or a regulator found an invented fact in something with our name on it tomorrow, could we show who checked it and when?

The Move

A federal judge learned this week that "my clerk used AI" is not a defense. Your company won't do any better with "our vendor used AI" or "the tool wrote that." If we signed it, we own it.

Ask the general counsel for a one-page rule before the next board meeting: anything AI drafts that leaves the building under our name has a named person who checked it against the source, and a record that they did.

The clerk can use any tool. Someone still has to read the draft.

Seven bullets. One move.