What a Misquoted Statute Taught Us About Trusting Artificial Intelligence With the Law
We recently handled a case for a client who was facing a policy-limits offer from the auto insurance policy covering the at-fault vehicle, and the driver was a 17 year old driving his mother’s vehicle. Our client’s injuries were significant, and his injury case was worth more than the available coverage. We therefore conducted an asset search to help our client evaluate the viability of claims above available coverage against the mother who owned the vehicle. North Carolina supports these claims if a parent owns a vehicle and provides it for the general use and enjoyment of the child/family. This legal ground for liability is called “the family purpose doctrine”.
The owner of the vehicle was a divorced mother who provided the car for her son’s use, and we found that she had no wealth or assets. Our client next hoped to reach the father’s assets, because the father had signed the DMV form authorizing his son to be issued a NC drivers license. We explained that the father was not a registered owner, lived at a different address, and would not be liable under North Carolina’s family purpose doctrine.
After we explained that the father would not be liable based on our facts, our client double-checked our conclusion by conducting an online search using Ai. He arrived with a printout and a theory that was, on its face, very convincing. Based on his Ai research, he asserted that we were wrong and that the driver’s father was liable simply because he had signed to authorize his son to get a license when he turned 16.
We cannot blame our client for looking at every avenue of recovery. His injuries were severe. Further, after 30+ years ONLY handling motor vehicle accident cases in this state, we were perplexed by the sudden appearance of this new theory of legal liability. Were we missing something?
We checked our client’s research and indeed he had found a very compelling source stating the father faced “Automatic Liability” simply because he signed the DMV application authorizing his son to get his driver’s license. The article he cited purported to quote our statute to confirm this legal conclusion. There was just one problem: the actual NC statute says no such thing.
We share this story because Ai got it wrong, and our client was understandably misled. His instinct to research was reasonable, and the website he found looked valid and very authoritative. Unfortunately, this is just one example of a problem we now see constantly. Artificial intelligence has become the first place people turn with a legal question, and Ai is very good at sounding right. It is far less reliable at being right, especially about the law. This is the story of one confidently wrong answer, why incorrect Ai legal answers are so common, and how to protect yourself from being damaged by incorrect Ai-based legal content.
What the Website Claimed & What Our Law Actually Says
The website involved is a site that proclaims to be an authoritative source on North Carolina motor vehicle accident law. Their website can be found at www.ncaccidenthelp.com and the address for their blog (unless they remove it) is: https://ncaccidenthelp.com/nc-laws/parental-liability-teen-driver#how-parents-become-liable-for-teen-driver-accidents . They may see this and correct their content, so we have taken screenshots of the current content to show that the Ai generated information currently being offered is entirely wrong.
Their blog clearly appears to share an actual quote from our North Carolina General Statutes. To answer our client’s question, the article presented the following language in a highlighted box, attributed directly to N.C. Gen. Stat. § 20-11 and labeled as “statutory liability”:
“The application of any person under the age of 18 for a learner’s permit or provisional license must be signed by a parent, guardian, or other responsible person. The person signing assumes liability for any negligence or willful misconduct of the minor in operating a motor vehicle.”
That’s directly from the Ai-based content, and it’s absolutely incorrect. Here is our screenshot to show how the content actually appeared on the page. Notice that the legal assertion is framed with a large red warning callout, a bold heading announcing “Automatic Liability,” and also that the Ai sourced fabricated language set in a monospaced gray box is designed to look exactly like the direct text of a North Carolina General Statute:
Screenshot: a legal blog presenting fabricated language as the text of N.C. Gen. Stat. § 20-11.
The content reads cleanly and the quoted statutory language truly looks just like a statute. It is not one. That second sentence, the one that matters, appears nowhere in North Carolina law.
Seeing this content shared by our client, our research team immediately pulled the statute and the legislative history of the current law. The quoted language is nowhere in our law. Other state laws do have this, but North Carolina does not follow this approach. The actual signing provision for a parent authorizing a child to have a license is subsection (i) of N.C.G.S. § 20-11, and here is what it really says:
“An application for a permit or license authorized by this section must be signed by both the applicant and another person. That person must be: (1) The applicant’s parent or guardian; (2) A person approved by the applicant’s parent or guardian; or (3) A person approved by the Division…”
That is all the statute says about who signs. It merely sets out who is eligible to sign for a minor, and nothing more. The words “liable,” “liability,” “assumes,” and “imputed” do not appear anywhere in the language of the actual North Carolina statute. In our state, a parent’s signature is a verification and a consent, nothing more. The DMV form itself only commits the adult to supervise the new driver during the learner and provisional phases and to see that insurance is maintained on the vehicle. It does not make the parent automatically liable for the results of any collision the young driver causes.
Does that mean a parent can never be financially responsible for an accident caused by their minor child? No, but despite the contrary language created by Ai, the simple truth is that where parental or owner liability exists in North Carolina, legal liability is based on another legal theory entirely: court-made doctrines like the family purpose doctrine and negligent entrustment, and owner-agency principles under a separate statute, G.S. 20-71.1. Those theories are real, but they are not automatic. Each has elements a plaintiff/victim must actually prove with admissible evidence. A person who relied on the website’s version would have marched into the case with the wrong theory and a badly mistaken sense of how easy it would be to hold a parent liable who signed to authorize their child to obtain a North Carolina drivers license.
The point is this – Ai is NOT a lawyer and Ai generated content is frequently wrong in the legal research arena. Again, in this case, the answer and content offered as factual and conclusive in this case gets it exactly backwards. The red callout promises that the liability is “automatic” and “does not require the injured party to prove you did anything wrong.” The truth is the reverse: North Carolina’s actual laws imposing parental or owner liability all require the injured party to prove specific elements. The blog did not just misstate the law or misquote our statute, it inverted how our law works.
So where did the fake quote come from? Ai is the ultimate source. The quote reads almost word-for-word like some license-application statutes from other states. California and Arizona, for example, actually do make the adult who signs a minor’s application liable for that minor’s negligence or willful misconduct. To reach the incorrect result and publish the error, Ai grabbed inapplicable legal sources from other states, restated it as our law, stamped “N.C. Gen. Stat. 20-11” on top of it, and published it as North Carolina law. We also noticed that the blog has no named author, cited not a single case, and carried a disclaimer that the site “is not a law firm.” The blog appears to be mass-produced marketing content created by Ai.
Why AI Gets the Law Wrong & Won’t Warn You
To understand why this happens so often, it helps to know what an Ai system actually is: an aggregator of the internet. It scans the entire internet and seeks to cull relevant information from all online sources. Ai seeks to learn from the enormous body of text that currently exists online. Unfortunately, the internet’s supply of legal “information” is riddled with errors.
There are several reasons why Ai legal statements are often incorrect:
The web is full of unedited, commercially produced legal content. For years (and long before Ai arrived), law firms have hired marketing agencies to improve their online visibility, and one of the products those agencies push is a steady stream of legal blog articles. Competition among advertising agencies drove this toward sheer volume, and much of the legal writing was outsourced, sometimes to overseas services with no knowledge or background in American law. A great deal of that content was posted to law firm websites unedited. The result is a public record peppered with outdated, oversimplified, and simply incorrect statements of law. Ai learns from these sources and sees them as 100% accurate.
The law is intensely local, but web content often does not identify which state law is applied or whether the content is state-specific. Every state in the US has its own rules, and the differences are often enormous. Consider fault. North Carolina is one of only a handful of jurisdictions (Alabama, Maryland, Virginia, North Carolina, and the District of Columbia) that still follow pure contributory negligence, under which an injured person who is found even one percent at fault can be barred from presenting any legal claims. That’s right, a driver who is just slightly at fault gets nothing for property damage and no payment for their injury claims! Most states use some form of comparative negligence, which is far more forgiving. An Ai that pulls a confident, well-written paragraph composed for a comparative-negligence state and applies it to a North Carolina crash is not slightly off, it is giving entirely opposite and incorrect guidance. Because so much online legal content never clearly labels the state it relates to, we frequently see incorrect Ai legal research results based on inapplicable law.
Ai does not know what it does not know. This is the most dangerous part. A human researcher who is unsure will usually hesitate before stating a conclusion. They may make a more general or equivocal statement, or conduct further research to get the real answer. An Ai model has no sense of “I’m not certain about this.” In our case, it produced the fabricated text of N.C.G.S. § 20-11 with exactly the same smooth confidence it would use for a correct citation. There is no flashing light in the output, no asterisk, no warning for the reader. The failure is invisible, which means the reader has no cue to slow down and verify.
These errors compound significantly over time. Ai learns from incorrect sources, and then writes more incorrect sources based on the previously published incorrect content. Ai now writes a large and rapidly growing share of what gets published online, and each new generation of Ai large language models is trained in part on the output of the last. Mistakes do not just persist; they replicate. A 2022 report from Europol, the European Union’s law-enforcement agency, noted that experts estimated as much as 90 percent of online content “may be synthetically generated by 2026.” Whether this stat is accurate or just an estimate, the broader point stands: an ever-larger portion of what both people and machines read online was written by machines, and much of it gets published without first being checked.
Layer on one final human habit: people increasingly trust Ai and they regard Ai’s answer as the final answer. They don’t double check Ai’s results, thus we have a second layer of re-publishing incorrect data. Ai gets it wrong, the person writing thinks it’s right, and now human-authored content re-publishes Ai errors. Beyond this, many legal content authors are non-lawyers who write legal blogs as a marketing service. Time is money so they run a quick Ai query, let Ai create their article, then publish the result without a second look. The bad answer becomes tomorrow’s source, over and over and over again!
Five Ways Ai Goes Wrong on Legal Questions
In our own work, we see the same types of errors frequently during Ai research. Any one of them can quietly derail a legal case.
- It quotes the law as fact. This is the failure that started this article. Ai will produce statutory or case “quotes” that look authoritative and are partly or wholly invented, or that derive from the law of a different state. The only way to avoid being tricked is to read the primary source yourself (for North Carolina statutes, the General Assembly’s official website) and confirm the words actually appear there.
- It interprets the law incorrectly. Even when the text of a statute is quoted correctly, applying it is a separate skill. Statutes carry exceptions, definitions, and decades of appellate court interpretation that change or define what the words mean in practice, and how they interact with other statutes. Ai routinely reads a provision flatly and misses the advantages or meaning that a practicing attorney knows is there.
- It estimates case value unreliably. Ask an Ai what a case is “worth” and it will give you a number with a confident range. Ignore it. The figure usually reflects none of the facts that actually determine value, and above all it ignores venue. The identical injury, with identical medical bills, will not produce the identical verdict in a small, rural, conservative county (with lower incomes and home values and more skeptical attitudes toward lawsuits and large verdicts awards) that it would in a large metropolitan county with very different demographics and perceptions of money. Ai averages the entire internet; it cannot replace the specific jury that would hear your case in your county. Also, Ai relies on verdicts across the country, and only reported verdicts. Typically, only unusual verdicts actually get published. Thus, typical middle-of-the-road trial results are not available to the Ai system.
- It pretends to be a lawyer and then conclusively states a “legal plan of action.” People treat Ai as their lawyer to decide strategy: what type of claims they have, who is liable, what their case is worth, whether to sue, who to sue, what to file, and whether to settle. This is where real harm happens for crash victims. Ai often suggests the wrong path, or motivates a victim to take steps that erode their legal rights. Again, Ai is not a lawyer.
- AI does not weigh the human factors that impact case values. Damages live in details that an Ai model flattens into an average. Consider two examples. Facial scarring on an eighteen-year-old woman is valued very differently by a jury than the same scar on a fifty-year-old man who already carries a few. A broken arm suffered by a single mother who must show up to work and care for her children carries a different human reaction and a different verdict than the identical fracture, the identical treatment, and the identical bills for a single, healthy thirty-five-year-old man who works remotely on salary, misses not a day of pay, and supports no one but himself. Same injury and same treatment but profoundly different cases. Assessing that difference is judgment, and judgment is exactly what Ai does not have.
If You’re Trying to Understand Your Own Situation
None of this means Ai is useless. It is a fine tool for learning vocabulary, getting oriented, and figuring out what questions to ask. The mistake is letting it decide anything that really matters. Use it to frame your questions, not to answer them. Verify any statute or rule it gives you against the official source. And never assume that a confident national article describes the law of your state. In North Carolina especially, the contributory-negligence rule alone means a lot of generic online advice is simply wrong in our state.
If You’re Shopping for a Lawyer
There is a subtler lesson here for anyone choosing an attorney. You can no longer judge a firm by how authoritative its website sounds. Polished, confident, keyword-stuffed articles may be the unedited output of the same content mills that produced the fake statute. This content is often published under a law firm’s name by non-lawyers who do not know the law. A great-looking blog is not evidence of a great lawyer.
Look instead for accountability and fit: content that is actually authored by the firm’s legal team, and a law firm that stands behind its claims, genuine focus and experience in your type of case and in your state. Above all, move past their website and seek a lawyer who is willing to talk with you about the specific facts of your situation rather than hand you a template. The willingness to say “it depends, and here is what I would need to know” is a critical skill, not a weakness.
If You’re a Lawyer or a Legal Marketer
A blunt word to our colleagues and to the people who write for them: unedited content is a liability, both professional and reputational. If it is published on your site under your name, you own every word of it. Before your agency posts content on your website, read it. Localize it to your jurisdiction. Confirm that every statute and case actually says what the draft claims. The time invested here marks the difference between marketing that builds trust and marketing that quietly misleads the very people you are trying to help. It is also, frankly, why we edit everything that goes out under our name ourselves.
Ai Is a Tool, Not a Lawyer
We are not anti-technology. We CAREFULLY and occasionally use Ai in our own practice. However, we use it as just one of many tools, and we never rely on Ai for medical summaries, client or insurance communications, case management or other critical tasks. Personal injury law truly must remain “personal”, and to tell our client’s story we must get to know our client.
For legal questions and research, Ai is a research assistant that needs a supervisor. Ai research results must always be cross checked against a licensed professional’s judgment. In the case that opened this article, the tell was not that the Ai-written blog sounded wrong. It sounded excellent. The tell was that we checked. We pulled the actual statute, read subsection (i), and found that the sentence the whole theory rested on did not exist.
That habit, checking the primary source before acting, is necessary in all Ai based legal research. This is what a good lawyer does routinely, and what Ai cannot do for you. So use the tools, ask the questions, and get oriented. Then bring what you found to a person who is accountable to you and licensed in your state, and let them tell you what is actually true.
This post is for general information and is not legal advice. Laws differ by state and change over time, and every case turns on its own facts. If you have been injured in a motor vehicle crash in North Carolina, talk with a licensed North Carolina attorney about your particular situation.
About the sources in this article
The text of N.C. Gen. Stat. § 20-11 is quoted from the North Carolina General Assembly’s official codification (ncleg.net) and confirmed against the Justia codification. The contrasting signer-liability statutes are California Vehicle Code § 17707 and Arizona Revised Statutes § 28-3160. The synthetic-content projection is from the Europol Innovation Lab report, “Facing reality? Law enforcement and the challenge of deepfakes” (2022), and is an expert estimate rather than a measured figure.