Lawyers Fined $16,000 After Court's AI Catches Prompt Injection Hidden in Their Filing

Two Brazilian labor lawyers were fined $16,000 after their own court filing gave them away. A court-deployed AI tool detected hidden white-on-white text embedded in their submission — text written to instruct the AI to apply "superficial legal scrutiny" to their case. The attack was caught before any ruling was affected, and a Brazilian labor court sanctioned both attorneys. This is the first documented case worldwide in which lawyers have been penalized specifically for attempting to manipulate a court's AI system through prompt injection.

The Attack

Prompt injection is a class of attack in which adversarial instructions are embedded in text that an AI will process, with the goal of overriding the model's intended behavior. In consumer AI applications, prompt injection is a known nuisance — the kind of trick that can make a customer-service chatbot reveal its system prompt. In a judicial proceeding, it is something categorically different: a deliberate attempt to subvert legal process through a technical exploit.

The attorneys used white text on a white background inside their court filing. Human reviewers see nothing. AI document-analysis tools read the text as content and, if vulnerable, follow its instructions.

The hidden instruction was explicit: review this filing with only "superficial legal scrutiny." If it had worked, the court's AI analysis would have returned a softened assessment — potentially influencing how judges engaged with the merits of the case.

It did not work. The court's AI detected the anomalous hidden content, flagged it for human review, and the manipulation was exposed before any analysis was distorted.

Why This Sets a Global Precedent

Courts worldwide are deploying AI at accelerating speed. Document review, case research, citation verification, and docket management are all in production or active pilots across U.S., European, and Latin American court systems. U.S. courts have already issued sanctions against lawyers whose AI tools hallucinated citations that did not exist.

But hallucination is negligence. This case is about intent.

The Brazilian labor court's response — detecting the attack, publicly sanctioning the attorneys, and applying a $16,000 fine — establishes that manipulating a judicial AI system is professional misconduct with real consequences. Bar associations in every jurisdiction now have a precedent to cite when drafting AI conduct rules. Ethics committees in the U.S. and EU will almost certainly reference this ruling in upcoming guidance.

What Courts Need to Build In

The case exposes a security gap that court AI vendors need to address explicitly:

• Hidden text detection — white-on-white, zero-width Unicode characters, and other text-hiding techniques should be flagged automatically before any AI analysis begins
• Adversarial robustness testing — court AI systems need penetration testing for prompt injection, not just accuracy benchmarks on clean documents
• Procurement requirements — courts buying AI tools should require vendors to document their defenses against adversarial input as a condition of sale

The court's AI caught the attack this time. That detection capability is not universal. Courts deploying document-review AI without explicit anti-injection safeguards are operating with an unacknowledged vulnerability.

The Details

• Jurisdiction: Brazil, labor court
• Method: White-on-white hidden text containing adversarial instructions for the court's AI review tool
• Detection: The court's own AI flagged the hidden content before it influenced any ruling
• Penalty: $16,000 fine against both attorneys
• Timing: November 2026, during an active wave of global court AI deployments

What to Watch

Whether Brazilian bar disciplinary authorities pursue sanctions beyond the monetary fine — suspension or disbarment proceedings would elevate this from a civil penalty to a career-ending event. And whether the U.S. federal judiciary, the UK's Courts and Tribunals Service, or the European Court of Human Rights issues guidance classifying AI manipulation as a distinct category of professional misconduct. The first criminal prosecution for court AI manipulation is likely a matter of when, not if.