Years ago, some clients would show up at my office with a file filled with their legal research. It would include pages of decisions from myriad jurisdictions which bore headnotes tangentially touching upon the issues in their case, but reflected their near-total ignorance of how the law actually worked.
They would demand that I read their research, which I told them I would happily do provided they would pay me for the time spent. Or they could just let me do my job and pay for that instead. Most would pick the latter. Those who refused tended to not be the sort of clients I wanted to represent, or the sort of clients who wanted a lawyer like me. Rarely were they willing to pay me to read their work rather than merely accept their representation that it was an “easy” case and they had already done all the work. This was just as well, as I was disinclined to waste my time or charge clients for work that was unproductive.
Now it’s judges who find themselves in that position, as artificial intelligence has emboldened clients to go it alone. Who needs a lawyer when you have ChatGPT?
Using ChatGPT and Claude, Mr. Sauve filed a new complaint in federal court. This time, it was a neatly typed document accompanied by 50 additional filings including a “case law synthesis” of legal research he said backed up his claim. In an interview, Mr. Sauve said A.I. had provided “the only path forward” for his case.
“Knowledge is power,” he said.
Federal judges and legal experts said they are increasingly seeing filings like Mr. Sauve’s flooding court dockets and clogging an already overburdened system as A.I. supercharges pro se litigation — even as it opens up the legal system to people who might not otherwise be able to afford to bring a case.
The pro se plaintiff lost, as did 96% of pro se litigants between 1998 and 2017. But is this really a problem, or an outlier?
But judges, lawyers and academics say the volume of filings by pro se litigants has risen dramatically alongside A.I.’s widespread adoption. The proportion of pro se cases filed by non-prisoners increased from 11 percent of all civil cases five years ago to 16.8 percent in 2025, according to a new study by two doctoral candidates that has not yet been peer reviewed.
The study found that much of the increase comes from the use of A.I. by pro se plaintiffs. The number of pro se complaints flagged as likely containing A.I.-generated text rose from virtually zero in 2019 to more than 18 percent in 2026, the study found.
But that’s a study. What about the reality on the ground?
Steven Donohue, a staff attorney for the United States District Court for the District of Minnesota in charge of reviewing pro se filings, said he observed a roughly 50 percent uptick in filings from non-prisoners starting around March 2025.
The problem is that in the past, pro se filings ranged from the crazy, hand-written nonsense to the occasionally really funny effort. Judge Kopf loved to tell about the “motion to go fuck yourself” he got from a pro se litigant. It was a welcome break from the tedium of lawyer filings. But ChatGPT has changed the nature of these submissions, which can now be cranked out with the quasi-appearance of actual motions and briefs, case citations and quotes (some of which are accurate, though many are not) and arguments that give the appearance of being substantive even when they never quite satisfy any cause of action.
The problem is that a judge, not to mention her clerks and/or court attorneys, has to read through this dump of filings, as their duty demanded.
“Judges still only have 24 hours in a day,” said Anand V. Shah of the Massachusetts Institute of Technology and one of the study’s authors. “Something has to give at some point.”
For litigants, the power of generative A.I. lies in its ability to turn a few short prompts from a user into lengthy documents with headers, citations and other earmarks of a legitimate legal brief.
For lawyers, this means that our papers, our cases, are being edged out by pro se litigant filings. There were problems in the past with getting timely decisions from the courts (New York Supreme Court Justice Edward Greenfield took eight years to decide a motion, which was a record at the time), but what’s a judge to do when his docket is engorged with pro se litigants filing 50 docs at a time filled with AI generated tripe and it pushes a decision in a critical motion that would normally be made in a week off for a month. Or a year?
Do we need more judges, more courtrooms and more clerks? Probably, but nature abhors a vacuum and the time will likely fill with more litigation rather than address courts being overwhelmed with AI generated pro se cases. Should judges just stop bothering with pro se litigants, or limit their reading to the primary papers and put the rest in the shredder? That’s unfair to the litigant who has a legitimate cause for redress and deserves to have his case heard and fully considered.
But the problem is only going to get worse as litigants realize they can crank out pseudo-legal docs at huge volume without a lawyer, and believe that the worst they can do is lose, but they might win, just like ChatGPT tells them they will.
*Tuesday Talk rules apply.
Discover more from Simple Justice
Subscribe to get the latest posts sent to your email.
Imagine believing that AI generated knowledge.
“Day after day they pour out speech;
night after night they communicate [… something].”
I believe it would only change the nature of the legal idiocracy if, for example, we tried executing board and C-level execs for every false citation a company’s generative AI produced. That’s a risk I’m willing to contemplate a bit longer, to be honest.
One wouldn’t want to deny access to the folks who really need to send [… something …] to today’s Judge Kopfs*. However, while I’m making modest proposals, perhaps minds or pseudo-legal agents could be sharpened by requirement of a rising, per-page bond for each complaint.
* As if there could be another
More likely than not, the courts will end up using AI to screen out the AI dreck.
[Ed. Note: Recently, I received a tee shirt from that concert from a kind reader. I love it.]
We need to see a photo of you wearing that tee while listening to the CDs of the show.
[Ed. Note: No one ever needs a photo of me.]
Point taken.
ChatGPT, using available online images produce a composite picture of Scott Greenfield of the blog Simple Justice wearing a tee-shirt from the Grateful Dead Live At Barton Hall, Cornell concert.
The clear solution is for pro se litigants to ask Claude to include whimsical humor with a mix of “go fuck yourself” as well as telling the court to STFU.
The first time I met Rich he said I’d recognize him by his ball cap. It was emblazoned with the letters “STFU,” which he opined stood for the “St Francis Fighting Utes.”
Would it kill you to make a like button?
When I started reading SJ, I had no clue how much I didn’t know. Luckily, you spared no time is disabusing me of that notion. Since then, I’ve had a chance to learn about what I had no f’ing clue about. The chance to hear and chat at length with Judge Kopf. The chance to expand my knowledge beyond that which the average pro se filer probably has in the tank.
When I chose to become a litigant for the first time, suing my local school board*, I didn’t yet have the benefit of AI, so I went in old-school. I downloaded briefs and opinions. I asked for advice here and there, and lo and behold – I won.
My second time as a litigant (again against my crappy Board of Ed), I have the benefit not only of AI, bt a good First Amendment attorney taking the case pro bono.
Looking at what Harvey created for an Amended Complaint vs what a real attorney created doesn’t make me worry that the legal field is in danger of AI taking over any time soon.
But, I will say this — there’s a lot of inequity built into the system. Inequities that would never be righted because no attorney would ever find them worth their time to litigate. To the extent that simple actions like these can be resolved in a litigant’s favor because they have access to AI, it’s a net plus. Perhaps not for the judge that needs to slog through it, but for the real help it can give to people.
There’s the old adage about if you judge a fish by its ability to climb a tree.
AI isn’t for the work you do. But that doesn’t mean it can’t be helpful in the legal field for a whole lot of people.
And it’s getting better. Which of course, means that roadblocks will be erected in due course (the judge hearing my Federal Action has a local rule requiring you indicate every word that resulted from the help of AI in your papers).
But who knows, maybe we’ll figure out some equilibrium.
* don’t do this, btw… it will either end with them winning or you winning and not getting anywhere near what you had hoped to accomplish.
No, no. You have an attorney. NOT an attorney + an LLM. Please just do what your attorney says and don’t send them slop. Be a good client and do as you are told. You haven’t a clue yet.
Trust me PK, I’m not substituting my LLM for her experience. And I didn’t send her the AI versions for any pointers.
But I still have access to the best AI that legal is using, so I figured I’d get a sense of what a side by side comparison looked like.
While my attorney was drafting the amended complaint, I asked Harvey to do the same.
My conclusion is that it’s… not there, yet.
Life’s but a walking shadow, a poor player
That struts and frets his hour upon the stage
And then is heard no more. It is a tale
Told by an idiot, full of sound and fury
Signifying nothing.
Macbeth Act V, Scene V 24-28
“New York Supreme Court Justice Edward Greenfield took eight years to decide a motion, which was a record at the time”
Does anyone actually believe he was deep in contemplation on this and not just a total lack of accountability and responsibility?
I’m somewhat sorry about this, but I can’t help myself…and it is Tuesday Talk.
As I read the post and the comments, I couldn’t help but wonder what an AI would provide as comment here. So I tried it with Microsoft’s CoPilot:
The funny part is that nothing about this is new except the word count. Pro se litigants have always believed they were one good citation away from vindication. Now they can generate fifty pages of that belief in under a minute. The volume changed, not the substance.
What has changed is the burden on the courts. A handwritten “motion to go fuck yourself” takes ten seconds to triage. A 40‑page AI‑brief with citations—half hallucinated, half irrelevant—forces a judge to pretend it might contain an actual claim. That pretense is the real cost.
People keep saying AI “opens access to justice,” but access to a keyboard was never the barrier. The barrier was understanding what a cause of action is, what jurisdiction is, what facts matter, and why “knowledge is power” doesn’t mean “I read a case from the Ninth Circuit that uses a word I like.”
The courts can’t ignore pro se filings, and they can’t hire clerks fast enough to keep up with the sludge. So the system absorbs the hit, and the paying litigants wait longer for decisions because someone’s 18‑page ChatGPT “case law synthesis” has to be skimmed for the one‑in‑a‑thousand chance it isn’t nonsense.
AI didn’t democratize the law. It democratized the illusion of competence. And illusions scale a lot faster than courts do.
(I think it’s a fail because it didn’t bring anything new to the discussion.)