Lawyer Lin Shang-lun’s article》Writing legal pleadings with AI as a layperson is digging your own grave

Generative AI has led courts to see a “pleading explosion.” Practicing attorney Lin Shang-lun said in a letter to the editor that pleadings produced by giving a case to ChatGPT or Gemini by a layperson are not only full of loopholes in the eyes of judges and opposing lawyers, but may also lead the parties to step into criminal liability for perjury and defamation. He specifically pointed out two hard flaws in the “web crawling to retrieve plus a vector database (RAG)” used by legal AI search tools on the market, and the most welcome opponent for professional lawyers is none other than laypeople who use AI-generated pleadings to sue.

(Background recap: Lin Shang-lun’s special article》Amplify and Accelerate: the real AI capabilities missed by 99% of legal practitioners）
(Background supplement: Lin Shang-lun’s special article》When text becomes a production line: the industrial revolution of AI in the legal industry）

Table of Contents

Toggle

• I. There is no absolute truth in law, but AI only wants to be a “yes-man” that “pleases you”
• II. From “perjury” to “defamation”
• III. Why even the “crawler-based database building” used by professional software still leads to mass delusions
• IV. The opposing side using AI to write pleadings is every professional lawyer’s favorite
• V. Don’t use your amateur level to challenge someone else’s professionalism

Recently, generative AI has become a mainstream trend across all industries. In the legal circle, the most striking phenomenon is the “pleading explosion” in court filings.

Many members of the public who encounter disputes start trying to act as lawyers themselves to save on legal fees. They toss case outlines into ChatGPT or Gemini, hit Enter, and a seemingly well-formed statement of claim or notice magically appears. These laypeople carry these AI-generated pleadings into court with confidence, believing they’ve obtained a “magic weapon” to fight professionals.

Up close in the front lines, you’ll see that this is a huge disaster. The way laypeople think they’re being smart—judges and opposing counsel see it as not only riddled with loopholes, but also pushing the parties themselves toward the brink of criminal wrongdoing. From a professional lawyer’s perspective, I want to discuss why laypeople who blindly trust AI to draft pleadings ultimately often turn it into a tragedy of self-destruction.

I. There is no absolute truth in law, but AI only wants to be a “yes-man” that “pleases you”

Many people have a fundamental misunderstanding about AI, thinking it is a “perfectly objective legal encyclopedia.” But in reality, the world of law has never been the kind of black-and-white natural science.

For any legal dispute, there are countless rooms for interpretation behind it. When you ask AI whether some conduct is “right” or “can be sued” directly with a blank page—without providing any authoritative practice viewpoints—AI’s underlying logic is not to help you find objective legal truth. Instead, it tries to “read your tone” and give you the answer you most want to hear.

This “people-pleasing personality” is a feature of AI. When you ask it with a lot of resentment—“Is what the other side did fraud?”—to match your emotions, AI will follow your logic, using extremely professional legal terminology, to fabricate a set of prosecution reasons that looks airtight but is completely detached from reality. You think you’ve received legal validation; what you actually just did was join in a collective self-hype with a highly socialized “yes-man.”

II. From “perjury” to “defamation”

Many people think that using AI to write pleadings is at worst that you can’t win the case—your filing gets rejected by the judge. But in fact, to help you assemble the elements of a crime or tort claim, AI will very likely “add ingredients automatically” and “invent out of thin air” some factual details that you never even mentioned in the first place. Not only can it lead to losing the case, but it also creates the risk of criminal prosecution:

1. When submitting to the court: committing the crime of “perjury”

If you sign and submit to the court or prosecutor’s office the AI-invented details (for example: that the other side threatened you on the spot that they would hurt you, or that the other side knew it was counterfeit and intentionally set a trap), then once the prosecutor finds that the facts are totally inconsistent with reality, you will have committed perjury—making false statements about the other party to allege their commission of a crime. This is direct deception of judicial authorities.

2. When posted online: involved in “defamation and public insult”

This is even more common in practice. Many laypeople can’t stand it; not only do they send the AI-drafted notice of claim to court, they also tack that “embroidered and exaggerated” prosecution draft or statement right onto Facebook, Threads, or a whistleblowing community for “online public trial.”

Once you publish these AI-invented, unverified derogatory statements on a public internet platform, the other side can immediately turn around and sue you for defamation and public insult. Then when you argue in court that “it was all written by AI, I only copied and pasted,” it won’t mean anything—no judge will just shrug you off. Because the person who pressed the “publish” button is you, and you must bear criminal responsibility for spreading false statements.

III. Why even the “crawler-based database building” used by professional software still leads to mass delusions

At this point, someone will say: “Then I won’t use ordinary ChatGPT—I’ll use those AI search tools on the market that claim to be made specifically for legal professionals. Surely that won’t have problems, right?”

Unfortunately, the answer is still: it will still make mistakes, and it will be wrong in a way that’s hard to detect.

Most of the so-called “legal AI search tools” on the market currently use the technology of “web crawling to retrieve plus a vector database (RAG).” This technology has two hard-to-overcome flaws:

2. Context discontinuity: Legal judgments often run to several thousand or even ten thousand words. For these software systems to make AI able to read them, they use crawlers to bulk download the judgments from the Judicial Yuan, then “slice” them into countless small fragments to store in the database. When you search, the AI only pulls out a few “apparently related” fragments and then uses its weaker model to piece together an answer. This “cherry-picking from chapters” approach often puts the reasons from case A into the conclusion of case B, producing serious logical mix-ups.
3. Outdated information and reproduction of infringement: A self-built database always has a time lag and cannot be updated in real time. What’s more, the large-scale crawling and storing of the Judicial Yuan database raises ongoing compliance concerns in law regarding “reproduction of copyrighted edited works.”

That’s why professional legal AI teams have shifted to the latest MCP technology. MCP’s strength is that it “doesn’t store, doesn’t crawl, and doesn’t slice.” When we send a query to the AI, MCP is like an instant official gateway, allowing the AI to connect to the Judicial Yuan’s official website in a safe and compliant way to read the latest and most complete judgments word-for-word. Because it reads the 100% real “genuine official original text,” the AI doesn’t need to guess out of thin air—so it can achieve zero hallucination and zero errors.

IV. The opposing side using AI to write pleadings is every professional lawyer’s favorite

In litigation practice, when we discover that the opposing layperson is filing lawsuits with pleadings written by AI, we usually feel relieved. That’s because it means the other side has proactively handed us the flaws that are destined to lose.

The logical chain in AI-written pleadings often can’t stand up to scrutiny. We carry out a careful “AI leak-check while reading the record”:

2. Verify the Supreme Court case numbers it cites—more often than not, they are AI-fabricated “hallucinated judgments.”
3. Break down the facts it asserts, and you’ll find it is packed with unreasonable, entirely objectively unsubstantiated fictional episodes.

We only need to calmly point out these “invented case numbers” and “fabricated facts” to the judge in court, and the party’s credibility will drop dramatically. Not only does it make the other side lose on the spot, it also provides an entry point for our subsequent counterclaims for perjury and requests for civil damages.

V. Don’t use your amateur level to challenge someone else’s professionalism

AI certainly lowers the barrier for many technical tasks, but it has never lowered the value of “professionalism.” Especially in legal areas where the tolerance for error is zero and it can instantly affect a person’s freedom and property, blindly using consumer-grade AI to draft pleadings is no different from running around half-naked.

Today’s AI is like the most powerful all-purpose wand in《Harry Potter》. You can use it to cast powerful magic, but the prerequisite is that you must first learn how to recite the incantations and how to summon the magic power within you. It’s only a tool, not a “Dumbledore” or “Professor McGonagall” that can directly help you defeat Voldemort.

If you’re a legal layperson and don’t even understand basic litigation and defense tactics, the rules of evidence, or practical legal logic, yet you still imagine you can wave this wand wildly in court, the outcome will only be that your own incantations rebound, and you’ll be blasted away on the spot. In this battlefield where legal tolerance for error is extremely low and where it can instantly affect freedom and property, blindly using consumer-grade AI to draft pleadings is a standard self-sabotage that leads to criminal wrongdoing.

In the AI era, lawyers who know how to leverage cutting-edge technologies like MCP and use AI as a professional amplifier will benefit significantly; while laypeople who try to take shortcuts with AI to save a little money will most likely end up paying a painful price for their ignorance—right in court.

Frequently Asked Questions

What legal risks are involved with lawsuits written with AI?

The author points out two types of criminal liability risk. Signing and submitting AI-invented factual details to the court or prosecutor’s office could constitute the crime of perjury; posting an indictment draft containing AI-invented content on Facebook, Threads, or a whistleblowing community allows the other party to turn around and sue for defamation and public insult.

Why do AI search tools that claim to be made specifically for legal professionals still make mistakes?

The author points out that most tools on the market use “web crawling to retrieve plus a vector database (RAG),” slicing judgment documents into stored fragments. During querying, the AI only pulls out seemingly relevant fragments and stitches together an answer, which easily leads to applying the reasons from case A to the conclusion of case B; and a self-built database has a time lag plus compliance concerns about reproduction of infringement.