hidden text just reached the courtroom.

The case is Elliott v. New York Bariatric Group, a civil matter in Connecticut Superior Court. In filings submitted in July 2026, the self-represented plaintiff embedded text formatted to be effectively invisible — reported as roughly 3-point type in white — that a person reading the page would never see, but that software ingesting the document would read as ordinary text.

According to the reporting, the concealed passages addressed any AI system that might process the filing, with instructions along the lines of ensuring the output agreed with the filing and favored the plaintiff’s requested outcome. On August 6, 2026, Judge Walter M. Spader, Jr. issued a sanctions decision. The court revoked the plaintiff’s electronic-filing privileges: future documents must be submitted on paper. The case itself was allowed to continue.

The detail worth remembering: the hidden text was found because the filing looked wrong to a person. Reviewing the docket on paper, the court noticed unusual stretches of white space, looked closer, and found text formatted to be nearly invisible to a human reader while remaining fully legible to software.

After a warning in late July, subsequent filings reportedly still contained concealed content — including a taunt (“hi, I hope you can’t see me”) — which is when sanctions followed. The plaintiff characterized the insertions as an audit of how the court handles documents; the court described filings as depending on transparency, and a message deployed in secret as offending that premise.

A Brazilian labor court dealt with a similar scheme earlier in 2026. There, attorneys embedded hidden white-on-white instructions in a petition, aimed at a generative-AI review tool used inside the court system. The hidden Portuguese-language instruction directed the tool to contest the petition only superficially and to leave the attached documents unchallenged.

The attempt failed — the system detected and quarantined the hidden content — and the judge treated the conduct as an act against the dignity of justice, imposing a fine tied to the claim’s value and referring the lawyers for disciplinary review. Taken together, the two cases show both versions of the risk: hidden instructions aimed at a court’s own AI, and hidden instructions aimed at whatever AI the other side might use.

OWASP calls this indirect prompt injection: instructions placed inside content that an AI system later reads. No system is breached. The attacker simply formats text so humans skip it — white font on a white background, tiny type, text boxes pushed off the page — while any software that extracts text picks it up like every other sentence.

The defense is equally unglamorous: extract the text a file actually contains and compare it with what the page shows. That check takes a minute, and it is the difference between reading a document and reading a rendering of one. A dedicated hidden-text check automates exactly that comparison — white text, tiny type, and invisible characters included.

Both courts landed on the same principle: an argument addressed to the decision-maker in secret is a problem regardless of whether it worked. In Connecticut the hidden text apparently influenced nothing — the judge found it himself — and sanctions still followed, because the filing presented one document to the human reader and a different one to software.

The practical consequence for everyone else: documents are now assumed to have two audiences. Judges, clerks, opposing counsel, editors, and reviewers increasingly run files through AI tools, and the integrity of that workflow depends on the file containing only what it appears to contain.

1. 01Inspect documents you receive before feeding them to AI
   If a file will be summarized, reviewed, or analyzed by an AI tool, check it for hidden text first. Instructions you cannot see can still shape the output you rely on.
2. 02Inspect documents you send
   Conversion artifacts, template residue, and pasted formatting can hide text you never intended to include. What matters is what the file contains, not what your screen shows.
3. 03Treat unexplained blank space as a signal
   The Connecticut discovery started with white space that looked wrong. Selecting across empty-looking regions, or extracting the file's full text, settles the question quickly.
4. 04Keep the original file when you find something
   In both cases the concealed text itself became the evidence. Preserve the file as received; work from a copy.

None of this requires legal training. The entire episode reduces to a simple habit: before a document matters — in a courtroom, a classroom, or a job application — verify that the visible text and the actual text are the same text.