Three years ago, I wouldn't believe this next sentence. A recent Connecticut Superior Court sanctions order adds an important item to every litigator’s review checklist: inspect filings and discovery materials for invisible text designed to manipulate artificial intelligence tools. In Elliott v. New York Bariatric Group, LLC, a self-represented plaintiff embedded tiny, white-on-white instructions directing any AI system analyzing his filings to favor his position. The court responded by revoking his electronic-filing privileges while preserving his ability to file on paper.
The decision takes a measured approach to both self-representation and AI. Although courts traditionally afford latitude to pro se litigants, that flexibility does not extend to conduct the court views as intentional deception, especially after an explicit warning. The Court wrote: “A self-represented party is entitled to a degree of latitude in the form of their filings, and the Court reads them generously, looking past inartfulness to the substance the litigant is trying to convey. That latitude, however, carries a limit.”
The order also does not treat AI itself as the problem. Judge Walter M. Spader Jr. recognized that AI can improve access to justice and assist legal research, translation, and review, while emphasizing that professional judgment must remain human, independent, and verifiable. The Court wrote: “[t]he Court welcomes the plaintiff's (or any litigant's) use of artificial intelligence in preparing filings” and “[t]he same qualities that make these tools useful make them dangerous to the careless and available to the dishonest.”
Most significant is the court’s analogy to an improper ex parte communication. Hidden instructions intended to influence a court-facing analytical tool operate outside the visible adversarial record: opposing counsel cannot see them, evaluate them, or respond. That framing places prompt injection within familiar principles of candor, procedural fairness, and the integrity of judicial decision-making—not merely cybersecurity or technological misuse.
The practical lesson extends beyond this case. Courts nationwide are adopting varied, and not always public, approaches to AI-assisted brief review and drafting. Litigators cannot assume they will know when a filing may be processed by AI. Teams should audit their own submissions and scrutinize opposing filings by examining document styles, extracted text, metadata, and PDF text layers for concealed or anomalous content.
