The plenary chamber of Brazil’s Superior Court of Justice in 2019 (cropped). Photo: Lucas Pricken / STJ (CC BY 2.0).
Lawyers in Brazil have again been caught trying to manipulate a court’s AI. They hid instructions in an appeal telling the AI to write a “clear, firm and conclusive” statement that their appeal should be granted in full. The Superior Court of Justice fined them about $4,000 Tuesday, saying the attempt didn’t affect the outcome. The same court had already reported two similar attempts in September.
The court developed its own AI system, STJ Logos, to help employees analyze appeals and prepare drafts. On June 9, after earlier attempts to manipulate court AI, Brazil’s National Council of Justice approved security guidance urging courts to check filings for hidden instructions, test whether their AI can be tricked and train staff to recognize attacks. The guidance also called for preserving suspicious material as evidence.
- A Connecticut plaintiff who tried the same thing lost his electronic filing privileges. (Harris Beach Murtha)
Cyber
Fairfax City Hall in 2010. Photo: Idawriter / Wikimedia Commons (CC BY-SA 3.0).
AI helped researchers find and fix serious weaknesses in software used to run websites, prompting precautionary shutdowns of some government services in the U.S. and at least eight other countries this week. The weaknesses had existed for more than a decade in PHP, a programming language used to run websites. Security firm Wiz said AI helped one researcher demonstrate working attacks in a week, work that previously would have taken a specialist team months. The findings helped establish the danger so developers could address it. Developers released fixes Wednesday, October 21, and several affected agencies reported no evidence of a breach.
In Other News
A federal judge was removed from a case over AI-invented facts in his order. (Bloomberg Law; court opinion)
A Louisiana appeals court is building its own AI tools to check filings and organize records. ([sch]Legal Tech)
Municipalities should own AI data centers like broadband networks, one infrastructure adviser argues. (Benton Institute)
Dallas garbage trucks are scanning homes for code violations. (NBC 5)
A judge left a nine-year prison sentence in place for a man who raised nearly $250 million by falsely promising AI-powered trading. (Court filing)
An investigation examined security failures in Flock’s surveillance network. (The Washington Post)
Derbyshire, a county in central England, has another police officer under investigation over AI use. (BBC)
British police switched categories of law to reopen a sexual deepfake case involving a popular broadcaster and online personality. (The Bureau)
FEMA now lets local governments use cybersecurity grants to pay for a formerly free threat-sharing service. (GovTech)
A federal judge will decide what remedies to order, if any, after finding that the Office of Personnel Management improperly gave DOGE, the U.S. government-efficiency initiative, access to workers’ personal records in early 2025. (FedScoop)
About Plain Thinking
Grant Morgan is an attorney, former Cuyahoga County assistant prosecutor and former Tribune reporter.
Plain Thinking covers AI and consequential technology affecting lawyers, police and courts.
Every Tuesday and Friday · Subscribe · Contact · LinkedIn
.jpg){kind=link}
