Recording, Spyware, and Using AI in Your Divorce

From The Mid-South Divorce Guide, chapters 30 through 31, by
William W. Jones IV, a Memphis family law attorney and Rule 31 Listed
Family Mediator licensed in Tennessee and Mississippi. This guide covers
Shelby, Tipton and Fayette Counties in Tennessee and DeSoto, Tate and
Tunica Counties in Mississippi.

Questions this page answers

• Can I record my spouse in Tennessee?
• Is it illegal to read my spouse’s email?
• Can ChatGPT conversations be used in court?
• Is AI legal advice reliable for divorce?

Recording, Snooping, and
Spyware

This is the chapter that keeps people out of jail, so read it before
you do something you cannot undo.

Recording conversations

Tennessee is a one-party consent state. Under Tennessee Code
Annotated section 39-13-601, it is generally lawful to record a
conversation you are a party to. If you are on the call, you can record
the call.

It is a crime to intercept a conversation you are
not part of. Putting a recorder in a room to capture your spouse talking
to somebody else is a violation of section 39-13-601, punished under
section 39-13-602 as a Class D felony, which under
Tennessee’s sentencing statute carries two to twelve years and a fine up
to five thousand dollars. Section 39-13-603 also creates a civil cause
of action, so the person you recorded can sue you.

There is a parental monitoring carve-out for monitoring your minor
child’s internet use on a device you own. There is no
spousal exception. Being married to somebody does not give you the right
to intercept their communications, and courts have rejected that
argument repeatedly.

Federal law adds another layer. The federal Wiretap Act, 18 U.S.C.
section 2511, prohibits intentional interception of wire, oral, and
electronic communications, and most federal circuits have rejected the
interspousal exception people try to claim.

If your case crosses into Mississippi or involves a party in another
state, check that state’s law. Some states require all parties to
consent.

Reading email and
accessing accounts

The federal Stored Communications Act, 18 U.S.C. sections 2701 and
following, makes it unlawful to access stored electronic communications
without authorization. That covers logging into your spouse’s email,
their cloud storage, or their social media without permission.

Knowing the password is not the same as having authorization. This is
the mistake people make. Your spouse told you their password four years
ago so you could check a flight. That is not standing consent to read
their email in 2026.

Spyware and tracking

Installing monitoring software on your spouse’s phone or computer.
GPS trackers on a car titled to them. Keyloggers. Camera apps. Location
sharing they do not know about.

These range from civilly actionable to felonious depending on what
was installed, on whose device, and what it captured. And the consumer
spyware industry has an atrocious security record, with major products
breached repeatedly, which means the data you collected and your
identity as the purchaser can end up in a public leak.

Two things that
are legal and still hurt you

Recording a conversation you are part of is lawful in Tennessee. That
does not make every recording a good idea.

Holding your phone up in somebody’s face in the middle of an
argument is a bad look. Even when the recording captures
exactly what you hoped it would capture. A judge watching that video
does not see a careful person documenting misconduct. A judge sees
somebody who decided, in the middle of a fight in front of their own
children, that the most important thing happening was building a case.
And the person on the other end of that camera usually looks like the
one being provoked, because you provoked them.

Recording your children is worse. Sitting a child
down and asking them questions about the other parent, on video, is the
single fastest way to convert a case about the other parent into a case
about you. And it never sounds the way the parent thought it sounded. On
the recording you can hear the prompting. You can hear the leading
question. You can hear the child looking to the parent for the right
answer, and then giving it. What the parent thought was the child
finally telling the truth reads on playback as a child being coached,
and once a judge hears that, everything else you have said about that
child gets discounted.

I have seen a parent lose a custody position they should have won on
the strength of a recording they made themselves and volunteered into
evidence.

The point is this. Evidence you were sure would be your strongest can
be the thing that beats you, and how you got it matters as much as what
is on it. Before you start recording anything, call your lawyer and
ask.

Why this backfires even
when it works

Say you get the recording. Say it proves exactly what you
thought.

Now you walk into court with evidence obtained by committing a crime.
The evidence may be excluded. Your lawyer may have ethical obligations
that limit what can be done with it. You may face criminal exposure and
a civil suit. And you have handed your spouse’s lawyer the best possible
distraction from whatever the recording showed, because the case is now
about what you did.

I have watched a client with a genuinely strong custody position hand
it away by installing an app on a phone.

The line

You can look at what is yours and what is jointly yours. Joint
account statements. Mail addressed to both of you. Documents in shared
spaces of the home. A shared computer you both use.

You cannot break into what is theirs. Their individual accounts,
their phone, their separate email, their private devices.

When you are not sure which side of the line you are on, call your
lawyer before you do it, not after. This is a five minute phone call
that prevents a felony.

How Not to Use AI in Your
Divorce

This chapter did not exist in any divorce guide five years ago. It is
now one of the most important chapters in this book, because artificial
intelligence is doing more damage to family law cases than any
technology since text messaging.

Let me start with the sentence that matters most.

The chatbot does not know it is wrong, and it will not tell
you when it is.

Mistake
one. Do not let AI write your legal documents.

AI language models generate text that looks like legal writing.
Confident, formatted, full of case citations. Some of those cases do not
exist.

This is not a rare glitch. Legal scholar Damien Charlotin maintains a
public database tracking court decisions where a judge found reliance on
AI-fabricated content. As of September 2026 it documented over two
thousand cases worldwide, including more than a hundred and twenty in
family law specifically. The database is public and it grows every
week.

It started with Mata v. Avianca, Inc., where in June
2023 a federal judge in New York sanctioned two lawyers who submitted a
brief citing six cases ChatGPT invented. They were fined and required to
write letters to the real judges whose names had been attached to fake
opinions.

It has gotten worse, and it has reached Tennessee. In Whiting
v. City of Athens, decided March 13, 2026, the Sixth Circuit
Court of Appeals sanctioned two Knoxville attorneys whose briefs
contained more than two dozen fabricated citations. The court ordered
them to pay the other side’s attorney’s fees, imposed double costs, and
levied a fifteen thousand dollar punitive fine on each attorney
individually. The court’s language is worth quoting, because it applies
to you as much as to a lawyer: no filing “should contain any citations,
whether provided by generative AI or any other source,” that the person
filing has not personally read and verified.

And here is the one that should stop you cold, because it is a
divorce case. In Shahid v. Esaam, decided by the
Georgia Court of Appeals on June 30, 2025, a husband’s attorney filed a
brief in a divorce dispute in which eleven of fifteen citations
were to cases that did not exist. It got worse. The trial
court’s own order relied on two of the fictitious cases. The appellate
court vacated the order, sent the case back, and fined the attorney.

Think about what that means. A family’s divorce was decided by an
order built partly on cases that were never written, by a court that had
no reason to think the citations were fake.

If a licensed attorney can be fined fifteen thousand dollars
for this, consider what happens to you. Pro se litigants now
make up the larger share of documented AI hallucination cases, over
eleven hundred of them in Charlotin’s database compared to eight hundred
involving lawyers. Judges are often more forgiving of an unrepresented
person than of a lawyer, because you did not take an oath of candor to
the court. But that leniency does not extend to your case. Once the fake
authority is stripped out of your motion, there is usually
nothing left, and you lose.

Mistake
two. Do not talk to a chatbot about anything you would not say in a
deposition.

There is no legal privilege for your conversations
with an AI chatbot. None.

That is no longer just my opinion, and here is the case.

In United States v. Heppner, Judge Jed Rakoff of the
Southern District of New York ruled from the bench on February 10, 2026
and issued a written opinion on February 17. It is a criminal fraud
prosecution, not a divorce, and I will come back to why that does not
save you. The defendant had learned he was under investigation. He took
what his lawyer had told him, worked through it with an AI assistant,
generated roughly thirty one documents laying out possible defense
strategies, and handed the results to his attorney. The government moved
to get all of it. He claimed attorney-client privilege.

He lost on every element. The AI is not a lawyer, and privilege
covers communications with a licensed professional who owes you
fiduciary duties. He had no reasonable expectation of confidentiality,
because the provider’s own privacy policy says it may use inputs and
disclose data. And he was not seeking legal advice from the AI, which is
what the privilege is for. The court rejected work product protection
too, because the documents were not prepared by counsel or at counsel’s
direction.

The judge did leave one door open. He suggested the answer might be
different if your lawyer directs the use of the tool, so that it
functions as the lawyer’s agent. That is the nuance the internet
summaries drop, and it is the only version of this that protects you,
which is why it belongs in a conversation with your attorney rather than
in your own browser at midnight.

Now, why a criminal fraud case governs your divorce. Nothing in that
reasoning depends on the kind of case. It depends on what the tool is
and what the provider’s terms say. Neither of those changes when the
caption says divorce.

The companies say the same thing. In July 2025, OpenAI’s chief
executive said publicly that “there’s no legal confidentiality for
users’ conversations” with ChatGPT, that if you talk to a therapist or a
lawyer or a doctor there is legal privilege for it, and that his company
“haven’t figured that out yet” for AI. He said OpenAI would be legally
required to produce those conversations if subpoenaed.

That is the company that makes the product telling you the product
has no privilege.

Then it stopped being theoretical. In litigation between The New York
Times and OpenAI, a federal magistrate judge ordered OpenAI to preserve
ChatGPT output logs, and in November 2025 ordered the production of
twenty million anonymized user conversations. In
January 2026 the district judge affirmed that order, holding that
ordinary ChatGPT users are non-parties whose privacy objection did not
defeat production.

Twenty million conversations, produced in a lawsuit that had nothing
to do with any of those users.

So understand what you are creating when you type into that box.
Every one of these is a record:

“How do I hide money before a divorce.” “Can my wife find out about
an offshore account.” “How do I make my husband look like an unfit
parent.” “What do I say so it looks like I make less money.” “How do I
get out of paying alimony.”

Your spouse’s attorney can serve discovery asking
whether you have used AI tools, what you asked, and for the export of
your conversation history. Most services let you download your entire
chat history in one file. If you have been using an AI as your
co-conspirator, you have written your own cross-examination.

Mistake three.
Do not use AI as your therapist.

Divorce is lonely and it is three in the morning and the chatbot is
right there and it is endlessly patient. I understand the appeal
completely.

But here is the difference. When you talk to a licensed therapist,
that conversation is protected by an evidentiary privilege. It is not
absolute, and in custody cases it can be pierced if you put your mental
health at issue, which is a conversation to have with your lawyer. But
the protection exists and it is real.

When you type the same words into a chatbot, you have none of that.
You have a text file on somebody’s server that can be subpoenaed,
produced, and read out loud.

The things people tell an AI at three in the morning are the exact
things that hurt worst in a custody case. Doubts about their own
parenting. Rage at their spouse. Thoughts about the children they would
never say out loud. Substance use. Despair.

Talk to a human being who has a license and a privilege. If cost is
the barrier, tell your lawyer, because there are options.

Mistake
four. Do not fabricate evidence with AI, and be ready for the
possibility that somebody else did.

AI can now generate convincing audio and video. This has already
reached family courts. In a widely reported United Kingdom custody case,
a father was accused of making violent threats based on an audio
recording. The recording had been manipulated to insert words he never
said. His lawyer caught it by examining the file’s metadata.

Two directions on this.

Do not do it. Fabricating evidence is fraud on the
court. It is criminal. It will end your case and possibly your freedom.
There is no version of this that works.

And if evidence appears that you know is false, do
not just say it is fake. Tell your lawyer immediately and preserve
everything. Authentication rules require the party offering evidence to
show it is what they claim it is, under Rule 901 of the Tennessee Rules
of Evidence and its federal counterpart. There is a proposed federal
rule addressing AI-generated evidence specifically, though as of the
spring of 2026 it had not been published for comment and the advisory
committee had scheduled further study. So for now the fight happens
under existing authentication law, and it happens through metadata,
forensic examination, and chain of custody. Original files. Not
screenshots.

Mistake
five. Do not trust AI to tell you what the law is.

Everything in Chapter 25 about
the 2018 relocation rewrite applies here. AI models are trained on a
large body of text that includes a great deal of outdated law, and they
cannot reliably tell you which parts are current.

Ask a chatbot about Tennessee relocation and there is a real chance
you get the pre-2018 substantially equal intervals framework, because
that test was written about extensively for fourteen years and the
current test has only existed since 2018.

Ask about alimony taxes and you may get the pre-2019 deduction rule,
because that was the law for seventy years and has been dead since
2019.

Ask about the Tennessee child support cap and you will probably get
ten thousand dollars a month, because that number is all over the
internet and it has been wrong since October 2021.

There is a second-order version of this problem worth naming, and it
showed up while I was writing this chapter.

A federal court has now ruled on almost exactly the question this
chapter is about, and I will get to the case in a moment. What is worth
noticing first is what happened to it afterward. Within weeks the
internet filled up with confident summaries of that ruling on pages that
look like law firm content, and the summaries flatten it. A narrow
criminal ruling gets restated as a sweeping rule about divorce cases.
The one real limitation the judge wrote into the opinion, which is the
part a client would actually want to know, disappears entirely.

That is the environment you are researching in. Not only invented
cases. Real cases, described wrong, in a confident voice, on a website
with a law firm’s logo at the top.

Where AI is actually useful

I am not a technophobe. I run a heavily automated practice and I use
these tools every day. Here is the line.

AI is safe when you already own the input and the output is
organizational.

Sorting your own bank statements into a spreadsheet. Building a
timeline from your own calendar and your own records. Summarizing a long
email thread you are a party to. Making a list of questions to ask your
lawyer. Drafting a household budget for your post-divorce life.
Explaining what a term like alimony in
solido means so you can follow the conversation in my office.
Turning your rambling four-page account of what happened into an
organized one page your lawyer can read in three minutes, which
genuinely saves you money.

AI is dangerous the moment the input becomes a request for
legal strategy or the output gets treated as legal advice, as evidence,
or as a court filing.

That is the whole rule. Use it to organize what you know. Do not use
it to decide what to do.

Lawyer Bill’s rule on this

Do not type anything into an AI that you would not want printed out,
handed to your spouse’s attorney, and read into the record.

If you have already been using AI to strategize about your divorce,
tell your lawyer. Today. Not because you are in trouble, but because we
need to know what is out there before somebody else finds it. I would
much rather hear it from you in my office than read it in a discovery
response.

Download the whole guide. All 35 chapters plus four blank client forms, complimentary, no email required. Download The Mid-South Divorce Guide (PDF)

Questions about your own situation? The Jones Law Firm, 5100 Poplar Ave, Suite 708, Memphis, TN 38137. Call (901) 761-5353.