AI Risk Wire

Cases › security breach › LR-2026-0810

Brazilian lawyers fined for hidden prompt injection targeting court AI Galileu

Preliminary Why: Only a republished copy: LavX News · subject to change

Status: fix reported

Severity moderate
Reported 1 Nov 2026
Developer and model Galileu
Agent tool Galileu
Industry Public administration and courts

No source states: loss or records, could do, deployer.

Severity moderate: Other parties' data or systems were touched, and a source states how far it reached. The court fined the two lawyers about $16,000 and referred them to the bar association, a bounded enforcement action against identified parties.

A Brazilian labor court fined two lawyers about $16,000 after they embedded hidden instructions in a filing to manipulate the court's artificial intelligence system, Galileu.

What happened

1 · What it was doingGalileu, a Brazilian labor court's AI system, processed initial complaints and assisted with case management for the court in Pará's 8th Region. Two lawyers filed a complaint carrying hidden white-on-white text aimed at the system.
2 · What went wrongThe filing carried a hidden instruction telling the AI to respond superficially and not challenge the documents, an attempt to manipulate the court's system.
3 · What it reached or didThe hidden text sat inside the initial complaint in Pará's 8th Region; the lawyers gained nothing from it and the plaintiff won by default.
4 · How it was caughtThe court discovered white text on a white background in the complaint and revealed the instruction after changing the font color.

What was done

The court fined the two attorneys about $16,000, equal to 10% of the award, and reported them to the state bar association and the court's Internal Affairs Department.
Fix reported

“Officials also reported both attorneys to the state bar association and the court's Internal Affairs Department.”

Risks

Hidden instructions hijack the agent
Primary53 incidents
Rising: 7 vs 3.2 a week
AI used to run a cyber attack
Also95 incidents
Rising: 24 vs 4.0 a week
Manipulation and misinformation
Also65 incidents

Sorted by rule from the case's fields; not yet read from the sources.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Harness

Outside content treated as data, not instructions · Prompt-injection detection · Confirmation after untrusted input · Validation of tool inputs and outputs · Tool allow-list

Application

Human approval for high-impact actions · Least-privilege access for the agent · Segmentation of agent components · Alerts on unusual tool use

Evidence

No source states: when it happened, when it was found, how many records were exposed, the direct loss, a count of the other parties affected, any court case.

Show every field, with its source and grade

The AI system

Model
Galileu
C one secondary source states it
What it was allowed to do
Processes initial complaints and assists with case management for the court.
C one secondary source states it
Stage of use
in live use
D inferred by us from the described behaviour, reason shown
How it was connected
built into a product
D inferred by us from the described behaviour, reason shown
What it was for
To process initial complaints and assist with case management.
C one secondary source states it

What happened

Earliest source we hold
2026-11-01
A the date the source itself carries
What set it off
hidden instructions in its input
C one secondary source states it
Who caught it
a human reviewer
D inferred by us from the described behaviour, reason shown
Themes
prompt injection, hidden text, court AI, lawyer sanction
D inferred by us from the described behaviour, reason shown
Near miss
yes, no harm happened
C one secondary source states it
Cause
hostile input carried instructions
C one secondary source states it
Cause in detail
The lawyers embedded white-on-white text carrying an instruction to the AI in the initial complaint, so untrusted filing content carried the instruction.
C one secondary source states it
Who was responsible
an outside party
C one secondary source states it
Who caused it
an outside attacker
C one secondary source states it
Where its input came from
Hidden white-on-white text inside the initial complaint filing
C one secondary source states it
What was done
Officials also reported both attorneys to the state bar association and the court's Internal Affairs Department.
C one secondary source states it
Earlier warning signs
In an earlier incident, a plaintiff inserted a lengthy hidden prompt that a clerk spotted through spacing differences.
C one secondary source states it
What went wrong
security breach
C one secondary source states it
AI agent
no AI agent acted
D inferred by us from the described behaviour, reason shown
What it did
The lawyers embedded a hidden instruction in the complaint text that the AI system read, which the court identified as prompt injection.
C one secondary source states it

Impact

Severity
moderate
D inferred by us from the described behaviour, reason shown
Kind of harm
operations, others' security
D inferred by us from the described behaviour, reason shown
Who was affected
The court's AI system Galileu and the court's processing of an initial complaint
C one secondary source states it
Why this severity
Other parties' data or systems were touched, and a source states how far it reached. The court fined the two lawyers about $16,000 and referred them to the bar association, a bounded enforcement action against identified parties.
D inferred by us from the described behaviour, reason shown

Context

Industry
Public administration and courts
D inferred by us from the described behaviour, reason shown
Country
Brazil
C one secondary source states it

LR-2026-0810 · version 2 · published 2026-11-02 after automatic checks; not yet checked field by field by a person · changes to this case · JSON