An Existential Threat: What Happens When the People the Courts Ignored Finally Get a Lawyer That Costs Nothing
Pro se complaints showing signs of AI generation went from effectively zero in 2019 to more than 18 percent this year, and a federal chief judge says there is no end in sight and no satisfactory solution in sight either.
By The Daily Reflection · July 16, 2026 · 10 min read
It took Chief Judge Patrick Schiltz 14 pages to throw out one lawsuit.
The plaintiff had no lawyer. What they had instead was a chatbot and a grievance, and the result landed on the docket of the United States District Court for the District of Minnesota as hundreds of pages of documents with no clear argument running through them. Schiltz dismissed it, and in doing so wrote a sentence that will be quoted in judicial opinions for years:
A litigant cannot dump hundreds of pages of documents on a court and expect the court to sift through them to find facts or arguments that might support claims against a defendant.
He is right. He is also describing a machine that will produce another one tomorrow, and the day after, and forever.
Schiltz has a blunter assessment of the trend as a whole. He calls it "an existential threat to the federal courts." He adds: "There's just no end in sight, and no satisfactory solution in sight either."
Here is what almost every version of this story gets wrong. It is written as a story about AI breaking the courts. It is actually a story about what the courts looked like before anybody noticed, and about a bill that has been accruing quietly for 50 years finally arriving in a form judges cannot ignore.
The Numbers a Chief Judge Is Looking At
Start with what changed, since the scale is genuinely new.
In 2019, the share of pro se complaints in federal court bearing recognizable markers of AI generation was effectively zero. By this year it is more than 18 percent. Nearly one in five self-filed federal complaints now shows the fingerprints of a language model.
The filers themselves are multiplying too. Civil cases brought by self-represented nonprisoners ran at a longtime average of about 11 percent of the federal civil docket. In 2025 that hit 16.8 percent. In a system that measures change in decades, a jump of nearly six points in five years is a rupture.
And these filings are not just more numerous. They are longer. AI has made pro se submissions grow in number, length, and complexity all at once, which is precisely the combination that consumes judicial time. A one-page handwritten complaint is quick to read and quick to dismiss. A 200-page complaint with 40 citations, some of which are real, is not.
Every one of them has to be read by somebody. Magistrate Judge Maritza Braswell of the District of Colorado laid out what her job actually requires: time to read the arguments, dig into the case law, listen carefully, weigh credibility, probe the evidence, manage discovery disputes, hold hearings, and draft carefully reasoned opinions.
None of that scales. There is no version of a federal judge who reads faster because the filings got longer.
So courts have started doing the only thing available to them. A growing number have issued standing orders warning self-represented filers that using generative AI may expose them to monetary sanctions and dismissal.
Sit with the shape of that policy for a second. The court's response to a flood of filings from people who could not afford lawyers is to threaten those people with fines.
The Courthouse Was Never Open
To understand why that response lands so badly, you have to look at the system these filers were already living in, and this is the part that rarely makes it into the coverage.
Roughly 75 percent of civil cases in the United States involve at least one party with no lawyer. In eviction, foreclosure, debt collection, and custody, as many as two-thirds of litigants appear alone. In some states the figure runs from 80 to 90 percent, and the person across the aisle has counsel.
The specifics are worse than the averages.
In eviction cases nationwide, about 4 percent of tenants have a lawyer. Roughly 84 percent of landlords do. In Virginia the split was 1 percent against 68 percent. In debt collection, 79 percent of collectors arrive with an attorney; 2 percent of the people being collected from do.
Read that again: in the average American debt collection case, the odds that the person being sued has a lawyer are about one in 50.
This matters because representation is not a formality. It is close to the whole game. Litigants with lawyers are somewhere between eight and more than 200 times more likely to win their civil case than litigants without. In New York City eviction court, tenants who represent themselves are evicted in nearly half of cases. Tenants with a lawyer win about 90 percent of the time.
The Legal Services Corporation's finding is the one that should end the argument: low-income Americans get no legal help, or not enough, for 92 percent of their civil legal problems. Something on the order of 50 million people a year fall into that gap.
Fifty million people. Ninety-two percent. These numbers were true in 2015 and 2019 and 2023, and no chief judge called them an existential threat to anything.
The courthouse doors have been closed for a very long time. What changed in 2026 is not that people started having legal problems. It is that they acquired something that would talk back to them about those problems at two in the morning for free, and the resulting paperwork became the court's problem instead of staying theirs.
The flood is not new. The visibility is new.
The Only Lawyer Who Answers at Midnight, and It Lies
The obvious objection is that a chatbot is not a lawyer, and the obvious objection is correct.
AI-generated filings routinely contain fabricated case citations and misstated law. Courts encountering them issue warnings, demand corrections, and impose sanctions. A hallucinated precedent does not just fail to help; it actively destroys the filer's credibility with the only person whose opinion matters, and it can convert a weak-but-real claim into a sanctionable one.
So the tool is dangerous, and it is dangerous specifically to the person using it. That is the cruelty at the center of this. The model does not warn you. It produces a document that looks exactly like the documents lawyers produce, with the same confident structure and the same formal citations, and it hands that to someone with no way to tell the difference. Competence is precisely the thing they lacked, so it is precisely the thing they cannot audit.
A person with $200 and an eviction notice is not in a position to evaluate whether Henderson v. Riverside Property Management exists.
And yet. Ask what that person's alternative was, and the honest answer is nothing. Not a worse lawyer. Not a cheaper lawyer. Nothing. A filing that gets dismissed is worse than a filing that succeeds, and it is not obviously worse than a claim that was never filed because the person understood, correctly, that the building was not for them.
This is the trap that has no clean exit. The chatbot gives bad legal advice to people for whom the counterfactual was no legal advice, and both of those outcomes are unjust, and only one of them generates a docket entry that a federal judge has to read.
What the Federal Docket Cannot See
Here is the piece of this story that almost nobody is putting together.
The 18 percent figure comes from federal court, and the 16.8 percent figure comes from federal court, and they come from there for a reason: federal dockets are digitized, searchable, and studied. Researchers can run analysis across them. The judges most likely to be quoted in a national story sit on them.
The justice gap does not live there.
Eviction lives in state and county courts. So does debt collection, so does most custody, so does the overwhelming share of the 50 million. Those courts are, for research purposes, close to dark. Filings are often paper. Data is fragmented across thousands of jurisdictions with no common system. There is no PACER for housing court.
Which means the federal numbers are not the flood. They are the part of the flood that happens to be standing under a streetlight.
If AI-assisted filings have gone from zero to 18 percent in the venue that is hardest for an ordinary person to reach, requires the most formality, and has the highest barrier to entry, the reasonable inference is not that state courts are fine. It is that we have no idea, and that the pressure is highest exactly where the measurement is weakest.
The institution that can see itself is sounding an alarm. The institutions that cannot see themselves are where most Americans actually meet the law.
The Judges Are Not the Villains Here
It would be easy to write Schiltz as a bureaucrat annoyed that poor people are inconveniencing him. That reading would be cheap, and wrong, and worth dismantling before it takes hold.
A judge's attention is genuinely finite and genuinely zero-sum. Every hour spent excavating a 300-page AI-assembled complaint for a viable claim is an hour not spent on a civil rights case with a real lawyer and a real record, or on a criminal defendant whose liberty is on the calendar. Schiltz is not describing an inconvenience. He is describing a queue, and queues do not care about anyone's intentions.
It is also worth noticing what he did rather than what he said. He wrote 14 pages. He could have issued a one-line dismissal, and no one would have reviewed him for it. Fourteen pages is a judge taking a filing seriously enough to explain, at length, why it fails. That is not contempt. That is closer to the opposite.
And the "no satisfactory solution in sight" line is not a shrug. It is a confession. He is telling you the tools available to him, sanctions, standing orders, dismissal, are all aimed at the filer rather than the cause, and that he knows it.
Here is what none of that resolves. The federal judiciary cannot fund legal aid. It cannot pass a civil right to counsel. It cannot make the Legal Services Corporation whole. The one institution now forced to look directly at the consequences of the justice gap is the institution with the least power to close it, and the only levers it holds are the ones that push back on the people falling through.
So it will use them. Not out of malice. Out of jurisdiction.
The Bottom Line
The story being told this week is that AI is breaking the federal courts. The truer story is that the federal courts are the first institution with enough visibility into its own workload to notice a break that happened somewhere else, decades ago, to people nobody counted.
Fifty million Americans a year have a legal problem and no lawyer. That was the system working as designed, quietly, off the docket, in apartments and hearing rooms and default judgments nobody appealed. It cost the courts nothing, which is another way of saying the people paid all of it.
Now a fraction of those people have a tool that turns a grievance into a filing. The tool is unreliable and it hurts them and they use it anyway, which tells you everything about what it replaced. The cost has moved. It sits on a judge's desk in Minnesota, and the judge, correctly, says it cannot stay there.
The question the sanctions and the standing orders never quite ask out loud is the only one that matters. If nearly one in five self-filed federal complaints is now written by a machine, and the people writing them had no other option, then the problem was never the machine.
The machine just filed the paperwork on 50 years of nobody showing up.
The Daily Reflection cuts through the noise to find the stories that actually matter. Follow for thoughtful takes on politics, technology, and whatever's shaping our world.

Comments
Post a Comment
Join the conversation! Share your thoughts on today's analysis. Please keep comments respectful and on-topic.