Portuguese 
Spanish 
Cases registered in SĂŁo Paulo, ParĂĄ, Minas Gerais, and other states reveal attempts to influence AI systems used by the courts and expand the debate on digital security in the Judiciary
A new concern has begun to gain space within the Brazilian Judiciary.
Hidden commands inserted in legal petitions were recently identified by courts in different states. The sections remained invisible during human reading but could be interpreted by artificial intelligence systems used to support case analysis.
Cases registered in SĂŁo Paulo, ParĂĄ, and Minas Gerais have expanded the debate on the risks of the so-called prompt injection, a technique used to insert instructions capable of influencing responses produced by AI tools.
-
World’s Largest Particle Accelerator Begins Four-Year Hiatus for Magnet Overhaul, Aiming to Hunt Dark Matter by 2030
-
“Sahara Dust Turns Rain Red, Startling Residents Across Europe”
-
Stanford Scientists Develop 20-Cent Paper Centrifuge Inspired by Ancient Toy, Spins at 125,000 RPM Without Electricity, Separates Plasma in 1.5 Minutes, Aiding Diagnostics in Remote Areas
-
World’s Largest Vaccine Manufacturer, Immunizing Two-Thirds of Global Infants, Originated in a Pune Horse Farm
Experts assess that the problem may be larger than the episodes already publicly identified.
Discovery in SĂŁo Paulo raises alert in the courts
One of the most relevant cases was identified by the SĂŁo Paulo Court of Justice.
During the analysis of a petition filed in 2025 against a financial institution, a hidden text containing instructions directed at artificial intelligence was found.
The command requested that the tool grant free legal aid, approve any urgent relief, and summon the defendant.
The discovery led Judge Diego Marcussi to request clarifications from the lawyer responsible for the document on May 19, 2025.
In the magistrate’s assessment, the presence of the hidden section could represent an attempt to influence technological systems potentially used in support of judicial activities.
According to the SĂŁo Paulo Court of Justice itself, the identification occurred during the supervised use of the artificial intelligence tools adopted by the institution.
Similar cases appear in other states
Similar situations were also recorded in different regions of the country.
In ParĂĄ, the Regional Labor Court of the 8th Region identified a hidden command in a case analyzed in the city of Parauapebas.
The instruction directed the artificial intelligence to contest a certain petition superficially and not to question the documents attached to the case.
As a consequence, two lawyers were fined R$ 84,200.
Minas Gerais also entered the list of states that reported similar occurrences.
On May 29, 2025, a judge from the Ibirité district imposed a fine of R$ 8,100 on a lawyer after identifying a hidden command in a case involving Banco BMG.
A few days earlier, on May 20, 2025, the Superior Court of Justice announced the opening of an internal investigation to investigate possible fraud attempts related to the use of technological systems.
Experts say the problem may be just the beginning
The recent episodes have led experts to advocate for a broader discussion on the use of artificial intelligence in the Justice system.
Dierle Nunes, an associate professor at the Federal University of Minas Gerais, states that the known cases may represent just the âtip of the icebergâ.
According to the researcher, many similar attempts may not have been identified so far.
The concern is growing because the use of artificial intelligence has become increasingly common in Brazilian courts.
A survey from the research Artificial Intelligence in the Brazilian Judiciary, coordinated by Dierle Nunes and Minister Luis Felipe SalomĂŁo, indicated that about 60% of the courts were already using some type of AI in 2025.
Manipulation can occur in various ways
The concern of experts goes far beyond texts written in white font.
Attached documents, supplementary files, external links, and jurisprudence databases can also serve as channels for manipulation attempts.
Researchers also warn of more sophisticated techniques that use specific word combinations to statistically increase the chances of artificial intelligence producing a certain response.
This scenario creates new challenges for courts investing in the technological modernization of their systems.
Experts advocate for mechanisms capable of filtering, verifying, and validating information before it is processed by automated platforms.
CNJ reinforces rules for the use of artificial intelligence
The National Council of Justice already establishes rules for the use of these technologies.
The rules mandate mandatory human supervision and prohibit judicial decisions made exclusively by automated systems.
New initiatives have also begun to be discussed.
According to information presented by the CNJ, studies, national research, and awareness campaigns are being developed to guide the proper use of artificial intelligence in the legal environment.
Courts have also begun to discuss protection strategies against hidden commands and other forms of interference in the systems.
Artificial intelligence will continue to advance in Justice
Experts do not advocate for the interruption of the use of artificial intelligence in courts.
Technological tools already help to speed up analyses, organize documents, and increase the efficiency of judicial activities.
Challenges related to security, transparency, and control, however, accompany this process of digital transformation.
Professional training, permanent human review, and advanced protection mechanisms appear as essential factors to ensure that technology continues to support the proper application of Justice.
Do you believe that artificial intelligence is ready to take on increasingly important roles in courts, or will human supervision remain indispensable for many years?
