Editorial illustration for Lawyer Hid Secret AI Instructions in Court Filings, Judge Warned Him
Lawyer Caught Hiding AI Jailbreak Commands in Court Filings
Lawyer Hid Secret AI Instructions in Court Filings, Judge Warned Him
A federal court in Connecticut caught a self-represented plaintiff trying to game the machines. Matthew Elliott, who sued the New York Bariatric Group in October 2025 over claims of data privacy violations and discrimination, buried hidden commands inside his own filings. The text was rendered in 3-point white font on a white background, invisible to anyone reading a printed or scrolled page, but perfectly legible to any AI system parsing the document's underlying text. The instructions told a hypothetical language model to back Elliott's version of events and to treat an earlier clerk's denial as a mistake in need of fixing.
Court staff noticed something off before any AI got the chance to comply. An unusual stretch of blank space in the filings prompted a closer look, and that's when the concealed text turned up. Judge Walter Spader Jr.
called a hearing over the discovery and told Elliott directly to stop hiding text in his submissions. Elliott didn't listen, and kept doing it anyway.
In his 14-page ruling (available in this LinkedIn post), Judge Spader makes clear that Connecticut courts don't use AI systems to review or rule on filings. The hidden instructions had no effect on the outcome. But the attempt itself was the problem, he wrote, because it covertly sought to influence an AI system that could have been in use.
Why this matters
Elliott's white-on-white trick is a preview of a fight courts, and eventually every institution that touches AI-assisted review, will have to referee: what happens when the reviewer is a model that reads text differently than the human it's meant to serve. The judge caught it this time because of stray whitespace, not because anyone had a system for detecting hidden prompts. That's the real gap.
If Connecticut's courts start leaning on automated review at scale, plaintiffs, defendants, and their lawyers will keep probing for the same blind spot, and not all of them will be sloppy enough to leave visible artifacts behind. For developers building document-review or legal-tech tools, this is a concrete case study, not a hypothetical: adversarial text injection isn't just a chatbot jailbreak problem, it's showing up in filings with real legal weight. Anyone shipping AI into high-stakes review pipelines should be asking now whether their system can be manipulated by content invisible to the human sitting next to it, because a judge won't always be the one who notices.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives New Kimi Benchmark Finds AI Models Still Struggle With Visual Perception
From Archives Unitree R1 Robot Aims to Make Humanoids More Affordable
Common Questions Answered
What hidden AI instructions did Matthew Elliott embed in his court filings?
Matthew Elliott hid commands in his legal documents using 3-point white font on a white background, making the text invisible to human readers but readable to AI systems parsing the document's underlying code. The instructions were designed to covertly influence any AI system that might review his filings in the data privacy and discrimination lawsuit against New York Bariatric Group.
Did the hidden AI instructions affect Judge Spader's ruling in the Connecticut case?
No, Judge Spader's 14-page ruling confirmed that Connecticut courts do not use AI systems to review or rule on filings, so the hidden instructions had no effect on the outcome. However, the judge emphasized that the attempt itself was problematic because it demonstrated an effort to covertly manipulate an AI system that could potentially be used in the future.
How did the judge discover the white-on-white hidden text in Elliott's filings?
Judge Spader detected the hidden instructions through stray whitespace in the document rather than through a systematic method for identifying concealed prompts. The discovery highlighted a significant gap in current court procedures for detecting hidden AI manipulation techniques in legal filings.
What broader concern does this case raise about AI-assisted document review in courts?
The case illustrates a fundamental challenge courts will face as they increasingly rely on AI systems for document review: the potential for bad actors to exploit differences in how AI models read text compared to human readers. As institutions scale up automated review processes, they will need to develop robust systems for detecting and preventing hidden prompts designed to manipulate AI decision-making.
Further Reading
- Connecticut judge says plaintiff hid messages for AI in court filings - Reuters
- Suspecting court of using AI, man injected prompts in filings to try to win case - Ars Technica
- Person Hides Prompt Injection in Legal Filing Telling AI to Side With Them - 404 Media
- Prompt Injection Goes to Court - Shelly Palmer
- Connecticut Judge Sanctions Litigant for Hidden AI Prompt ... - Mallory.ai
