November 19, 2026
A line of tiny white type in a document can try to tell an AI summarizer how to frame its conclusions. Two courts have already seen it.
In Connecticut this summer, a self-represented plaintiff embedded hidden white-on-white text in court filings directing any AI system that reviewed them to produce output in the plaintiff’s favor. The court does not use AI to review filings, and the judge read a printed copy. The court still revoked the plaintiff’s electronic filing privileges and compared the hidden text to an ex parte communication.
In Brazil, a labor court in Pará fined two lawyers for a petition containing a hidden line instructing AI to contest the document only superficially and not challenge the supporting materials. The judge noted the target could have been either the court’s tools or the opposing party’s.
Security researchers call this indirect prompt injection: instructions that reach an AI tool through content the tool is asked to read.
The same hidden line can sit in an email attachment, a vendor proposal, an expert report, or documents produced in discovery. In eDiscovery, that raises a practical question: does the processing and review protocol flag text a person cannot see but the tool can?
A court has a judge, a clerk, and a record. In a mediation or negotiation, a draft agreement or settlement proposal often goes straight to the other side — and may go straight into the AI tool they use to summarize it.
The trick is technical. The risk is to trust. Each side relies on what the other sends. That reliance is tested when part of a document is written for the other side’s tools rather than for the people using them.
Sources
Superior Court of Connecticut, “Elliott v. New York Bariatric Group, LLC, Memorandum of Decision” (August 6, 2026).
Dechert, “Connecticut Court Sanctions Litigant for Hidden ‘Prompt-Injection’ Text In Filings” (September 30, 2026).
3ª Vara do Trabalho de Parauapebas (via Migalhas), “Decision, ATOrd 0001062-55.2025.5.08.0130” (May 12, 2026).
Migalhas, “Juiz multa advogadas que esconderam prompt para enganar IA da Justiça” (May 2026).
OWASP Gen AI Security Project, “LLM01:2025 Prompt Injection” (2025).
Ariadne Latorre is a Florida Supreme Court certified mediator and a software engineer. She mediates business and commercial, insurance and risk allocation, LegalTech and information governance, HOA and community, and court-ordered matters, in English and Spanish, in Miami and virtually worldwide.