Skip to Content
logologo
AI Incident Database
Open TwitterOpen RSS FeedOpen FacebookOpen LinkedInOpen GitHub
Open Menu
Donate
Discover
Submit
  • Welcome to the AIID
  • Discover Incidents
  • Spatial View
  • Table View
  • List view
  • Entities
  • Taxonomies
  • Submit Incident Reports
  • Submission Leaderboard
  • Blog
  • AI News Digest
  • Risk Checklists
  • Random Incident
  • Sign Up
Collapse
Discover
Submit
  • Welcome to the AIID
  • Discover Incidents
  • Spatial View
  • Table View
  • List view
  • Entities
  • Taxonomies
  • Submit Incident Reports
  • Submission Leaderboard
  • Blog
  • AI News Digest
  • Risk Checklists
  • Random Incident
  • Sign Up
Collapse

Report 7314

Associated Incidents

Incident 14971 Report
Hidden Prompt Injection in Brazilian Labor-Court Petition Reportedly Tried to Manipulate Galileu

Loading...
Galileo: System identifies attempted manipulation in petition and alerts judge.
trt4.jus.br · 2026

Galileu, an artificial intelligence tool developed by the Regional Labor Court of the 4th Region (TRT-RS) and nationalized by the Superior Council of Labor Justice (CSJT), identified an attempted manipulation in the initial petition of a case judged by the 3rd Labor Court of Parauapebas (PA). The case is addressed in a ruling issued this Wednesday (May 12th) by Judge Luiz Carlos de Araújo Santos Junior.

Upon processing the document, Galileu detected hidden sections with instructions addressed to the artificial intelligence itself. The content instructed the system to contest the petition superficially and not to challenge the documents, regardless of the command it received. This technique is known as prompt injection. Upon detection, the system issued a prominent alert to the user, identifying the technical details of the occurrence. Furthermore, it prevented the maliciously inserted content from being processed by the tool.

Technical Report and Judicial Decision

Galileo limited itself to reporting the technical facts, without qualifying the conduct or proposing any procedural course of action. The magistrate did not decide solely based on the alert: he examined the content indicated by the system before taking any action. Only after this human verification did the judge assess the situation, decide on any potential consequences, and justify his decision, in accordance with the requirement for human review in the use of artificial intelligence by the Judiciary.

According to the Secretary-General of Technology and Innovation of the TRT-RS (Regional Labor Court of Rio Grande do Sul), Natacha Moraes de Oliveira, the case shows the importance of users prioritizing institutional tools that take all precautions, as Galileo did, when using artificial intelligence. "Controlling this type of attack, such as injecting commands or even data in an attempt to manipulate the results generated by AI tools, is not so trivial. It requires the application of specialized techniques to identify them," emphasizes Natacha.

Security by Design

Galileo's behavior—identifying the anomaly, alerting the user, and preserving human decision-making—complies with the main international technical guidelines on security in artificial intelligence systems, which recommend precisely this combination of measures in the face of prompt injection attempts (OWASP LLM01; NIST AI 600-1 MS-2.7-007/MS-4.2-001; NCSC/CISA §1; MITRE ATLAS AML.T0051).

The use of Galileo observes Resolution No. 615/2025 of the National Council of Justice (CNJ), which regulates the use of artificial intelligence by the Judiciary based on principles such as effective human supervision, information security, transparency, and respect for fundamental rights. End of news article body.

Source: General Secretariat of Technology and Innovation

Read the Source

Research

  • Defining an “AI Incident”
  • Defining an “AI Incident Response”
  • Database Roadmap
  • Related Work
  • Download Complete Database

Project and Community

  • About
  • Contact and Follow
  • Apps and Summaries
  • Editor’s Guide

Incidents

  • All Incidents in List Form
  • Flagged Incidents
  • Submission Queue
  • Classifications View
  • Taxonomies

2026 - AI Incident Database

  • Terms of use
  • Privacy Policy
  • Open twitterOpen githubOpen rssOpen facebookOpen linkedin
  • 638678f