[Administrative Attorney Fact-Finding ⑤] Why Fact-Finding Matters More in the AI Era

Hello, this is Jean, an Administrative Attorney in Korea.

A business receives an unexpected inspection. A few days later, a notice of an administrative fine arrives. But the inspection report is wrong. It says, “Operating after 10:00 p.m.,” even though the business closed at 9:50 p.m. that day.

How do you prove that?

CCTV footage may be overwritten after a certain period. Employee statements may be viewed as less persuasive because the employees have an interest in the matter. At that point, someone may suggest, “Why not ask an Administrative Attorney about fact-finding?”

So you search for “Administrative Attorney fact-finding.”

But something feels strange.

There is very little material that systematically explains this topic. No academic papers. No Supreme Court precedents directly organizing the issue. Even statistics on the issuance of fact-confirmation certificates are not publicly available.

Fact-finding is one of the duties expressly mentioned in the Administrative Attorney Act, yet it has rarely been explained in a systematic way.

That is why I started this series.

In this final installment, I will look at the role fact-finding may play in the AI era.

The Real Scope of Fact-Finding Under the Administrative Attorney Act
Fact-Confirmation Certificates: Why Neither Lawyers nor Private Investigators Can Issue Them
How Fact-Confirmation Certificates Are Actually Used
Where Administrative Attorney Fact-Finding May Be Used
Why Administrative Attorney Fact-Finding Matters More in the AI Era

Special Edition: We Asked the Government Directly About Administrative Attorney Fact-Finding

A suspect submitted KakaoTalk messages and a cash-withdrawal record as evidence in a criminal case. The police relied on the materials and proceeded with the investigation.

Later, however, it turned out that both had been manipulated using AI.

What happened next was even more disturbing.

The actual victim ended up being prosecuted for false accusation.

This was a real case uncovered by the Daegu District Prosecutors’ Office in July 2026.
(Gyeongbuk Maeil, July 16, 2026, “AI-Manipulated KakaoTalk Messages and Cash Withdrawal Records… Victim Nearly Became Defendant in False Accusation Case”)

When I read about this case, one thought occurred to me.

In the future, we may face not only problems caused by the absence of evidence, but also problems caused by an abundance of evidence that looks so convincing that we have to explain what is real.

This change is also connected to the fact-confirmation work performed by Administrative Attorneys.

Fact-finding by an Administrative Attorney refers to confirming factual circumstances within the scope permitted by the Administrative Attorney Act and documenting the results.

There is, however, one important point I want to make clear before continuing.

As discussed in the special edition, the precise scope of “fact-finding” under Article 2(1)(7) of the Administrative Attorney Act is currently awaiting a formal statutory interpretation from the Ministry of Government Legislation.

Therefore, this article is not an invitation to commission fact-finding work under Article 2(1)(7).

The issue I want to address today is more fundamental.

Why might a record showing that someone actually verified a fact at a particular time become more important in the AI era?

There is also an even more important question.

If you do not document a fact now, will you still be able to verify it in the same condition after a dispute begins?

In many cases, the answer is no.

That is what I want to discuss in the final installment of this series.

Can AI-Generated Fake KakaoTalk Messages Really Turn the Actual Victim Into a Defendant?

The case mentioned above occurred at the Daegu District Prosecutors’ Office in July 2026.

Person A used Person B’s personal information to obtain a loan of approximately KRW 19 million.

After discovering what had happened, B filed a complaint against A.

Normally, you would expect an investigation to follow and the facts to be determined.

But the situation was reversed.

A used AI to fabricate KakaoTalk conversations and a cash-withdrawal record and submitted them to the police.

The materials looked convincing.

The police decided not to forward A’s case for prosecution, and B was instead prosecuted for false accusation.

Think about the situation from B’s perspective.

B never had those conversations.

B never received the money.

But the other party produced materials showing what looked like a transaction, and even the investigative authorities believed them.

Would simply saying, “That evidence is fake,” solve the problem?

No.

A person claiming that evidence is fabricated still needs a basis for that claim.

Only after the trial prosecutor conducted additional search and seizure procedures and digital forensic analysis was the manipulation discovered.

The prosecution against B was withdrawn, and A was detained and indicted on charges including perjury and obstruction of official duties by fraudulent means.

The truth was eventually uncovered.

But what matters is what happened in between.

Without additional verification, fabricated material could have solidified into something treated as fact.

In the past, the common question was:

Do you have evidence?

Now another question may need to follow.

How do we know that the evidence is authentic?

One of the biggest changes in the AI era is not that evidence disappears. It is that fabricated evidence can look real.

Courts Are Already Receiving “Precedents” That Do Not Exist

The problem of AI-generated fabrication is not limited to photos or KakaoTalk conversations.

Case law can also be fabricated.

In an April 2026 case before the Seoul Northern District Court, a plaintiff cited a Supreme Court case number in a court filing.
(Law Times, March 23, 2026, “Court Notes That Plaintiff’s Cited Case Number Was an AI-Generated Fake”)

But when checked, the case number did not exist.

It was a fabricated precedent generated by AI.

Courts including the Seoul High Court, Incheon District Court, and Seoul Southern District Court have also encountered submissions containing nonexistent case numbers or fabricated precedent.

Such false precedents have even appeared in documents prepared by legal professionals such as lawyers and labor attorneys.
(Law Times, April 1, 2026, “Judgments Note ‘Nonexistent Case Numbers’… Even Fifth-Year Lawyer Submits Fake Citation”)

AI can be wrong.

The real problem is that it does not always look wrong.

It can be wrong in a highly convincing way.

A case number appears.

A court name appears.

A holding or legal proposition may also appear.

To someone unfamiliar with the relevant field, it may look entirely legitimate.

That is why the Office of Court Administration of the Supreme Court took action.

From November 2025 through March 2026, a “Task Force on Responding to False Claims and Evidence Submitted Through AI Use” operated with eight judges and two lawyers.

The task force discussed measures such as disclosure of AI use, sanctions for false citations to statutes or precedents, and possible disciplinary referral of lawyers who submit such material without proper verification.
(Supreme Court of Korea, March 31, 2026, “Response Measures for False Statutory and Precedent Citations Using AI”)

Since February 20, 2026, the Judicial Information Disclosure Portal has also operated a function allowing users to verify potentially false case numbers.

Why are these measures becoming necessary?

The answer is surprisingly simple.

The speed at which information can be created is becoming faster than the speed at which it can be verified.

I also use AI when organizing materials.

It is extremely useful for comparing dozens of documents, organizing sentences, and building structure.

Using AI in professional work is becoming as basic a skill as using Excel.

But the more I use AI, the clearer one point becomes.

Final verification still needs to be performed by someone who understands the field.

AI can produce material that looks plausible even when the underlying source does not exist.

It does not reliably stop and say, “I could not verify this.”

That means professional expertise may no longer be measured simply by how much a person knows.

The ability to distinguish between what has been verified and what has not may become even more important.

And more broadly:

As AI lowers the cost of generating information, the value of verification increases.

Why Is It Becoming Harder to Say That a Single Photo Is Enough?

Photos and video are not immune from the same problem.

In May 2026, during the appellate proceedings involving former Minister of the Interior and Safety Lee Sang-min, the defense submitted CCTV footage enhanced using AI.
(JTV, April 20, 2026, “Who Conducted the Video Analysis? Prosecutor Challenges Defense Over AI-Processed CCTV”)

The prosecution argued that the footage had not gone through an accredited forensic examination and that neither the person conducting the analysis nor the methodology had been clearly identified.

Regardless of the ultimate outcome of that case, the change revealed by the dispute is clear.

In the past, the primary question about CCTV footage was:

What does the footage show?

Now another question may need to be added.

Is this the original footage? What processing has been applied to it?

Since May 2024, the Supreme Prosecutors’ Office has also been developing deepfake detection technology and has filed a patent application relating to synthetic voice detection.
(Korea Economic Daily, May 8, 2026, “Prosecutors Develop Technology to Detect Manipulation of Evidence and Deepfakes”)

If prosecutors are developing dedicated technology to detect manipulation, this is no longer merely an interesting technological issue.

Photos remain important evidence.

CCTV footage remains important as well.

But when time passes and the other party begins to challenge authenticity, it may become necessary to explain not only the photograph itself but also when it was taken, where it was taken, what was verified, and how it connects to the original source material.

Preserving a photograph and preserving the factual circumstances that existed at the time are not necessarily the same thing.

The Most Important Issue May Not Be AI, but Time

Facts have one important characteristic.

Many of them disappear as time passes.

Suppose a business is accused of violating permitted operating hours.

The owner says:

We closed at 9:50 p.m. that day.

At that moment, there may still be many things that can be checked.

CCTV footage may exist.

POS records may still be available.

Employees may still remember what happened.

The physical condition of the premises may also remain unchanged.

But what happens one or two months later?

CCTV footage may have been overwritten.

Original POS records may be harder to retrieve.

Employees’ memories may fade.

The premises or operating practices may also have changed.

Then someone asks:

Can you verify what the situation was at that time?

But once a fact has disappeared, it cannot simply be recreated.

This is one of the most important points in fact-confirmation work.

No matter how capable AI becomes, it cannot reopen a past physical scene.

It cannot recreate CCTV footage that no longer exists.

It cannot reconstruct the exact condition of a store at 9:50 p.m. on a particular day.

That is why the value of documentation is often highest while the relevant facts still exist, not after the dispute has already escalated.

Before moving out, the condition can still be checked. Months after moving out, the same condition may no longer be reproducible.

Is Taking Photos Before Moving Out Enough?

Suppose a tenant is leaving a commercial property.

There are a few small marks on the wall, but no major damage.

Most of the fixtures are in normal condition.

The tenant takes several photographs and returns the keys.

A few months later, the landlord says:

The premises were seriously damaged.

The landlord then demands repair costs.

The tenant produces photographs.

But the landlord responds:

How do we know when those photos were taken?

How do we know they were not edited?

How do we know those photos are even of this property?

Having no photographs can be a problem.

But having photographs does not necessarily end the dispute.

The issue eventually becomes:

What condition was the property actually in at that particular point in time?

In situations like this, it may be more important to think about what should be documented before a dispute begins rather than waiting until the dispute has already occurred.

A dispute may begin later, but evidence may disappear before the dispute starts.

What Happens If the Other Party Changes Their Story Later?

This issue is not limited to physical locations.

The same thing can happen during contractual relationships.

Suppose two people enter into a goodwill-transfer agreement or begin a business partnership.

At first, communication is easy.

We will proceed under these terms, correct?

Yes, that works.

There are KakaoTalk messages.

There is a contract.

But several months later, the relationship deteriorates.

One party says:

That is not what I meant.

At that point, what seemed simple becomes a dispute over memory and evidence.

Which KakaoTalk messages are original?

What materials existed when the contract was signed?

Which file existed at which point in time?

In the AI era, the situation may become even more complicated because generating text, creating images, and modifying files has become easier.

So the important issue is no longer only:

“Who is telling the truth?”

Another question matters:

“What existed before the dispute began?”

That is why fact-confirmation should not always be viewed only as something done after a dispute arises.

Some facts are meaningful only if they are confirmed before the problem occurs.

A record created before the other party changes their story starts from a different position than a record created afterward.

What About Visa and Immigration Matters?

In visa and immigration matters, is what appears on paper always the same as reality?

Unfortunately, no.

AI can read a lease agreement.

It can extract an address.

It can generate a checklist of required documents.

But can AI directly verify whether the person actually lives at the stated address?

It cannot.

The existence of a lease agreement and the existence of an actual residential relationship are separate factual questions.

This distinction existed even before AI.

But as document generation becomes easier, the difference may become even more visible.

A larger quantity of paperwork does not automatically prove reality.

That is why I believe one question may become increasingly important in administrative practice:

“Who verified the real-world fact underlying this document?”

AI can handle the first stages of the work very quickly.

It can classify records.

It can organize them chronologically.

It can draft documents.

But it cannot complete the final step of connecting documents to the physical world.

What was actually present at that location?

Did that person actually live at the address?

Was that item actually in that condition?

Did that record actually exist at that time?

These are different questions from data organization.

AI can create documents. But it cannot directly witness reality.

What Does a Fact-Confirmation Certificate Mean in the AI Era?

There is no need to treat fact-confirmation certificates as the opposite of AI.

AI should be used well.

It is excellent at organizing dozens of documents by date, identifying repeated phrases, and comparing inconsistencies across records.

The important question comes afterward.

Will we simply trust the output created by AI?

Or will we go back and check the original material?

Will we review the date of creation?

Will we examine how different files are connected?

Will we clearly identify what could not be verified?

I believe the role of professionals may gradually move away from simply being “people who prepare documents for others.”

If AI can generate a draft in seconds, that function alone becomes less distinctive.

What remains valuable is:

the ability to decide what needs to be verified,

the ability to distinguish original material from claims,

the ability to determine what additional verification is necessary when records conflict,

and most importantly,

the ability to distinguish between what has been verified and what has not.

If a fact-confirmation certificate is viewed simply as “a piece of paper,” this difference is easy to miss.

The important point is not merely the document itself.

It is:

what fact the document addresses, what source it relies on, and the scope of what was actually verified.

In the AI era, the value of people who know what needs to be verified may become greater than the value of people who simply know how to produce documents.

When Should Facts Be Confirmed?

If a fact may be difficult to verify in the same condition later, it may be worth considering whether it should be documented now.

Examples include:

When it is necessary to preserve what materials and terms actually existed at the time a contract was concluded
Before moving out or handing over a building or facility
When original records such as CCTV or POS data may disappear over time
When existing records should be organized before a relationship with the other party deteriorates
When factual circumstances relating to recordings, contracts, or transaction records may need to be documented

This does not mean that a fact-confirmation certificate is necessary in every such case.

In some situations, simply preserving the original material may be enough.

In others, notarization, forensic examination, or another method may be more appropriate.

The important question is not “What document should I issue?” but “What fact may disappear if I wait?”

You do not need to organize every document before asking for an initial review.

If you already have contracts, photographs, recordings, or transaction records, you may begin with the most important materials and determine whether the matter may fall within the scope of a fact-confirmation certificate under Article 2(1)(2) of the Administrative Attorney Act.

Do You Need to Establish Facts Relating to a Contract, Photograph, or Recording?

This Concludes the Administrative Attorney Fact-Finding Series

This series began with a simple question.

What exactly does “fact-finding” by an Administrative Attorney mean?

Following that question led to far more issues than I initially expected.

| Series | Main Topic | | --- | --- | | ① The Real Scope of Administrative Attorney Fact-Finding | The statutory structure of the Administrative Attorney Act and the legal position of fact-finding work | | ② Who Can Issue a Fact-Confirmation Certificate? | The legal basis and scope of fact-confirmation certificates under Article 2(1)(2) | | ③ How Are Fact-Confirmation Certificates Actually Used? | Goodwill transfers, partnerships, transcripts, administrative appeals, residence-related facts, and other examples | | ④ Where Can Administrative Attorney Fact-Finding Be Used? | Administrative fines, move-outs, pre-construction conditions, deliveries, actual residence, and other factual situations | | Special Edition — We Asked the Government Directly | Ministry of the Interior and Safety response, the interpretation issue under Article 2(1)(7), and a request for formal statutory interpretation | | ⑤ The Meaning of Fact-Finding in the AI Era | The value of factual verification and contemporaneous documentation in the age of generative AI |

There were also disappointing aspects.

In particular, the response from the Ministry of the Interior and Safety raised further questions, and the lack of a formal statutory interpretation from the Ministry of Government Legislation remains unfortunate.

Once the Ministry of Government Legislation issues its formal interpretation, I will return with a follow-up article rather than ending the discussion here.

If you have followed this series, you may now be wondering whether you should organize and preserve materials you currently have, such as contracts, photographs, recordings, or transaction records.

The standard is simple.

Will this fact still be verifiable in exactly the same way one month from now?

If the answer is no, it may be worth considering in advance how the relevant facts should be documented.

Administrative Attorney fact-confirmation work and fact-confirmation certificates may serve as useful tools depending on the circumstances.

I will continue to write about contracts, factual verification, and other issues that are often overlooked even though they arise frequently in everyday life.

Follow the KakaoTalk channel and the blog if you would like to receive future updates.