Elilton Alves Gouveia v. Meridian Financial Investments, LLC

D ISTRICT C OURT OF A PPEAL O F T HE STATE OF F LORIDA
FOURTH DISTRICT
ELILTON ALVES GOUVEIA,
Appellant,
v.
MERIDIAN FINANCIAL INVESTMENTS, LLC,
Appellee.
No. 4D2025-0843
[March 25, 2026]
Appeal from the Circuit Court for the Fifteenth Judicial Circuit, Palm
Beach 	County; 	Maxine 	Cheesman, 	Judge; 	L.T. 	Case 	No.
502020CA004951XXXXMB.
Elilton Alves Gouveia, Lexington, South Carolina, pro se.
Cory Zadanosky of Schwed Kahle & Kress, P.A., Palm Beach Gardens,
for appellee.
MAY , J.
The 	defendant 	in 	a 	contract 	dispute 	appeals 	a 	trial 	court 	order
enforcing a settlement agreement. 	The defendant argues the trial court
misapplied the settlement agreement’s terms in enforcing the agreement.
We disagree and affirm. 	We write to call attention to the defendant’s
apparent use of artificial intelligence in his briefing to this court.
The 	underlying 	case 	began 	with 	a 	contract 	dispute 	between 	two
companies. 	The plaintiff (appellee here) filed suit. 	The parties reached a
settlement agreement, which led to the trial court dismissing the case with
prejudice, but retaining jurisdiction to enforce the terms of the settlement
agreement.
Part 	of 	the 	settlement 	agreement 	provided 	for 	a 	forensic
accounting/audit of the defendant’s company. 	If that audit had “any
findings,” the defendant was required to pay the plaintiff up to $400,000.
The audit led to yet another dispute, causing the plaintiff to file a motion
to enforce the settlement agreement.

2
The trial court granted the amended motion to enforce the settlement
agreement. 	From this order the defendant appeals.
We have de novo review of an order interpreting a settlement agreement.
See Sakowitz v. Waterside Townhomes Cmty. Ass’n, Inc., 338 So. 3d 26,
28 (Fla. 3d DCA 2022).
The defendant argues the trial court misinterpreted the settlement
agreement’s 	terms 	when 	it 	enforced 	the 	settlement 	agreement. 	The
plaintiff responds the trial court correctly interpreted the agreement’s
terms. 	We agree with the plaintiff and affirm without elaborating on the
details of this appeal.
• 	AI Spotted
There once was a litigant pro se,
Who let an AI lead the way.
It briefed every claim,
Cited cases—by name,
That vanished by morning’s next day.
Limerick on Pro Se Parties Using Artificial Intelligence (on file with the
Fourth District Court of Appeal) (generated by ChatGPT 5.2).
It appears to us the defendant used a large language model (LLM)1 to
write 	his 	briefs. 	Popular 	LLMs 	include 	OpenAI’s 	ChatGPT, 	Google’s
Gemini, and Microsoft’s Copilot. See In re Kenney, 2025-0389 (La. App. 5
Cir. 10/23/25), 422 So. 3d 905, 912 n.5.
Technology, specifically artificial intelligence, is a marvel of the age we
live in. 	It is an important and productive tool, but left unchecked for
accuracy and legitimacy, it can be a plague upon the judicial system,
creating more problems than it solves, and resulting in violation of the
rules of appellate procedure. 	As Judge Forst reminded us:
An attempt to persuade a court or oppose an adversary by
relying 	on 	fake 	opinions 	is 	an 	abuse 	of 	the 	adversary
system [….] Many harms flow from the submission of fake
opinions [….] These include wasting the opposing party’s time
1 	LLMs 	“are 	[artificial 	intelligence] 	systems 	that 	aim 	to 	model 	language,
sometimes using millions or billions of parameters[.]” See LAURIE H ARRIS , CONG.
RSCH. S ERV ., IF12426, G ENERATIVE A RTIFICIAL INTELLIGENCE: OVERVIEW, ISSUES ,
AND CONSIDERATIONS FOR CONGRESS (2025).

3
and money in exposing the deception, taking the court’s time
from other important endeavors, and potential harm to the
reputation of judges and courts whose names are falsely
invoked 	as 	authors 	of 	the 	bogus 	opinions 	and 	to 	the
reputation of a party attributed with fictional conduct.
Goya v. Hayashida, 418 So. 3d 652, 655-56 (Fla. 4th DCA 2025) (citation
modified).2
The defendant’s briefs were replete with case citations that either do
not exist or fail to support the defendant’s arguments. 	Here, for example,
the defendant cites non-existent cases such as Dausch v. Crane, 448 So.
2d 613 (Fla. 4th DCA 1984). 	And the defendant’s citation to Bennett v.
NationsBank, 759 So. 2d 1215 (Fla. 5th DCA 2000) leads the reader to
Summers ex rel. Dawson v. St. Andrew’s Episcopal Sch., Inc., 759 So. 2d
1203, 1206 (Miss. 2000), a case discussing punitive damages in a tort
action, far removed from the contract issues involved in this case.
Other cases cited in both the initial and reply briefs exist but address
unrelated issues. See, e.g., Gross v. Lyons, 763 So. 2d 276, 277 (Fla. 2000)
(adopting into Florida law the indivisible injury rule to be applied when a
jury cannot apportion injury); Cohen v. Kravit Est. Buyers, Inc., 843 So. 2d
989 (Fla. 4th DCA 2003) (reversing summary judgment due to genuine
issues of material fact in existence); Mullins v. Kennelly, 847 So. 2d 1151
(Fla. 5th DCA 2003) (finding 57.105 fees unwarranted); Murphy v. Bay
Colony Prop. Owners Ass’n, 12 So. 3d 924 (Fla. 4th DCA 2009) (finding
error in the trial court’s dismissal of a case based on the merits); De Groot
v. Sheffield, 95 So. 2d 912, 916 (Fla. 1957) (discussing the quasi-judicial
proceedings of the Civil Service Board); and Broward Cnty. v. G.B.V. Int’l,
Ltd., 787 So. 2d 838, 845 (Fla. 2001) (discussing the writ of certiorari and
site plans/plat applications).
By this opinion, we put the defendant on notice that future unchecked
use 	of 	artificial 	intelligence 	in 	filings 	with 	this 	court 	may 	result 	in
sanctions for failure to comply with Florida Rule of Appellate Procedure
9.210(c).
Affirmed.
2 Separately, the Florida Third District Court of Appeal ordered a pro se appellant
to show cause why he should not be sanctioned for using fake and semi-fake
citations in his briefs. See Takefman v. Pickleball Club, LLC, 418 So. 3d 826, 827
(Fla. 3d DCA 2025).

4
CONNER and L OTT , JJ., concur.
LOTT , J., concurs separately with opinion.
LOTT , J., concurring.
I concur fully in the majority’s well-written opinion.
I write separately to highlight the need for prophylactic, rather than
remedial, solutions to the problem of improper use by pro se litigants of AI
chatbots.
That is not to minimize the well-recognized problems of improper use
of AI by attorneys, particularly where AI generates hallucinated or fake
authority that the attorney submits to the court without verification. 	But
courts have been properly and adequately responding to this problem by
using existing rules and tools to sanction attorneys who engage in this
improper 	conduct. 	That 	toolbox 	works 	well 	enough 	for 	attorneys.
Attorneys are repeat players in litigation. 	Sanction them, and they will
learn from it. 	Monetary sanctions imposed on attorneys, who tend to be
solvent, 	can 	make 	their 	adversaries 	whole 	for 	the 	time 	wasted 	by
misconduct. 	If they repeatedly disregard sanctions orders, more severe
discipline can be imposed by courts or state bars. 	Over time, I have no
doubt that courts’ consistent response will lessen the problem of improper
AI use by attorneys.
Pro se litigants, on the other hand, are usually not repeat players in the
court system. 	The case at hand is their case. 	They have little experience
or knowledge on how to litigate cases and how to, or not to, use tools like
generative AI in that litigation.
All this creates a problem for the courts, for at least three reasons.
First, remedial sanctions or warnings, like the one the Court rightly
imposes on Appellant today, do nothing to prevent the problem of the
continued use of AI by new pro se litigants who never received such
warnings.
Second, there is a seemingly endless deluge of AI-generated drivel
submitted by pro se litigants who have never received such warnings. 	I
will not bother to collect authority sanctioning it; it is ample. 	Even more
of it is dealt with in unpublished orders. Most commonly, the recalcitrant
litigant simply loses without court comment on the AI problem, which is
often the most economical way for a court to dispose of a given dispute.
Any judge on any bench right now understands the pervasiveness of the

5
problem.
Third, the AI-generated slop that pro se litigants serve up is a unique
sort of gruel. 	Unlike real lawyers, AI Chatbots, at least in their current
form, do not “think.” 	They make predictions about what words ought to
come next in response to a prompt that the user provides it.3 	This
technology is very good at sounding right, but less adept at being right,
especially where critical thought is required in creation of the content. Pro
se 	litigants, 	reasonably, 	often 	do 	not 	appreciate 	the 	distinction 	and,
lacking legal training, do not appreciate how or why a response might not
be right. 	But it sounds right, so they put it in their brief to see what
happens. 	And since the cost to generate the content is so low, they can
put in a lot of it. 	The opposing party and the court are left in the position
of breaking down why something that sounds right is not right, which
tends to consume more resources than parsing through a traditional pro
se appeal.
Pro se litigants of course cannot be faulted for using these tools. 	The
lack of affordable legal services has been a perennial problem in the courts
and legal professions. 	The problem is that AI Chatbots appear to the
untrained eye to be a solution. 	But unless cautiously and thoughtfully
wielded, they are no solution; they make the problem worse.
So in order to meaningfully solve the AI-slop problem, we need to get
pro se litigants to understand, up front, that blind reliance on a Chatbot
for legal assistance is not acceptable.
This is a much more difficult task than warning or sanctioning litigants
on the back end. 	I have not seen a perfect solution.
Some courts have implemented rules or standing orders requiring all
litigants, attorney and self-represented alike, to disclose the use of AI and
3 E.g., Snell v. United Specialty Ins. Co., 102 F.4th 1208, 1227 n.7 (11th Cir. 2024)
(Newsom, J., concurring) (“As I understand things, the LLM that underlies a user
interface like ChatGPT creates, in effect, a complex statistical ‘map’ of how people
use language—that, as machine-learning folks would say, is the model’s ‘objective
function.’ How does it do it? Well, to dumb it way down, drawing on its seemingly
bottomless reservoir of linguistic data, the model learns what words are most
likely to appear where, and which ones are most likely to precede or follow
others—and by doing so, it can make probabilistic, predictive judgments about
ordinary meaning and usage.”).

6
certify its accuracy.4 This is probably the right starting point, and I would
support adoption of such a requirement for this Court.
Chatbots are going 	to get better, and 	that’s going to make these
problems worse. 	The question now must be how to address them on the
front end.
* 	* 	*
Not final until disposition of timely-filed motion for rehearing.
4 E.g., Jim Ash, 11th and 17th Circuits Order Disclosure, Certification of AI Use in
Court 	Filings, 	The 	Florida 	Bar 	(Feb. 	9, 	2026) 	(online 	at
https://www.floridabar.org/the-florida-bar-news/11th-and-17th-circuits-order-
disclosure-certification-of-ai-use-in-court-filings/)