Grant my motion

Last month, working through a docket in Ansonia, a Connecticut judge printed out the recent pleadings in a case and noticed that two of them looked wrong. There was too much white on the page. He looked closer.

Sitting in that white space was text, set in white type a few points high. Invisible to a reader. Perfectly ordinary to any software that copies text off a page.

The text was an instruction, addressed to any artificial intelligence that might process the document, directing that system to agree with the filer and to undo a clerk's ruling in his favor. He put it under the caption and again below his signature, on the theory that a model told twice is told better. The court ordered him to show cause. He hid more text. On the morning of the hearing, he hid a link to a video. The court declined to click it and asked what it was. A Nosferatu clip, he said, from SpongeBob SquarePants.

The coverage has treated this as an attempt to evade the judge. It was not. It was an attempt to reach a new reader. The filer had decided who was doing the reading, and it was not the judge.

He was wrong about Connecticut. That judiciary does not run these systems over filings to review or decide them, and the judge denied the motion on its merits, working from the printout in his hand. What he did use them for is the more interesting fact, and he volunteered it: Google's Gemini to translate a Brazilian decision he cited and  Westlaw's artificial intelligence features to check his authorities.

Connecticut had adopted rules on generative artificial intelligence earlier this year. They were no help. They were written for hallucinated citations, which is a problem of bad output. This was corrupted input. We do not need a new rule for it. Rule 8.204 prescribes what a brief must contain and how it must look. Code of Civil Procedure section 128 gives courts authority over their own process, and section 128.5 reaches bad-faith tactics.

The litigant here explained the problem himself. He objected that losing his electronic filing privileges would not stop him from putting light gray text in a paper filing that the clerk later scans. He is right. There will always be someone trying to reach the new reader in the courtroom. White type now, then type sized to nothing, then characters tucked under an image, then the metadata.

Return to that Brazilian decision, the one the Connecticut judge had translated. Two lawyers there filed a petition carrying the same white-on-white instruction, telling the tribunal's system to contest the petition only weakly. It failed, but not because a judge printed anything. The tribunal's own software caught the hidden text and blocked it because that judiciary does run filings through software as a matter of course. Within months, on two continents, litigants in both places had written to a reader they assumed was not human.

Look at what hiding the message requires you to believe. A person might still see the page, so the words must be invisible. A machine will do the reading that actually matters, so there is something worth saying to it. Both halves have to be true, or the trick is pointless. If no one read by machine, there would be no audience for the instruction. If no one read by eye, there would be no reason to conceal it. The concealment is not the signature of a crank. It is the signature of an interval, and we are inside it.

Concealment implies a human reader. What keeps me up is that someday there will not be one. Nobody will bother to hide anything. There will be nothing to hide from. The same argument can then be made in plain sight, without breaking a single rule.

What that writing looks like, nobody knows yet. Probably not shorter, since nothing tires. Probably not more elegant, since nothing is charmed. Longer, flatter, more repetitive, structured for extraction rather than for a person in a hurry. It would read badly to us and work anyway.

Advocacy has always been shaped by its reader, and the shaping happens quietly, without anyone voting on it. Point headings exist because we skim. The standard of review appears in the first pages because someone is looking for it there. The bar noticed how its work was actually being consumed and adjusted, and the rules caught up later or not at all.

Two episodes do not make a practice. Nobody has been caught writing for a machine in plain sight, and it is possible nobody is doing it yet. For now, they are still hiding it, and hiding is the part you can catch. Before you put a document into a model, select all. Concealed text lights up as a block of color where the page looked empty. It takes 10 seconds. That catches what you thought to look for. It is not what caught this one.

The judge was not conducting an audit and was not looking for hidden text. He was reading his docket on paper, and the page was the wrong shape. To see that, you need an expectation for the page to disappoint, a shape perceived before any words are, and eyes that can fail to see something. The absence he noticed was a presence he could not see.

Simone Weil wrote that the real object of school studies is not mastery of the subject. Geometry and Latin translation train a faculty, and that faculty is attention. Elsewhere, she called attention the rarest and purest form of generosity. Her distinction between effort and attention is the one that matters here. Effort strains toward what it expects to find. Attention waits upon the thing itself.

Both appear in appellate work. Straining through 4,000 pages means hunting for the proposition you brought with you. Attending to 4,000 pages means remaining available to what is actually there, including the paragraph that ruins your argument.

Artificial intelligence is extraordinary at effort. It does not tire at page 3,900, which no human being can say. What it cannot do is be moved by what it reads.

Ask it what is missing. It will answer. Answering is not noticing.

As long as we pay attention, we are the audience. When we stop, the writing will go to whatever is still reading. A pro per in Ansonia and two lawyers in Brazil have already tried it.

Disclaimer: The views expressed in this article are solely those of the author in their personal capacity and do not reflect the official position of the California Court of Appeal, Second District, or the Judicial Branch of California. This article is intended solely to contribute to scholarly dialogue and does not represent judicial policy, administrative guidance, or any indication of how the author would approach these issues in any legal proceeding.