Fined $5,000 Because He’s a Lawyer. For Small Businesses, It Could Mean Bankruptcy

Stephen Aarons, a lawyer in New Mexico, was fined $5,000 for submitting a fabricated “witness testimony” created by ChatGPT to the court. In an appeal for a murder conviction, he had the AI generate statements from non-existent witnesses and fake police testimonies, submitting them without verification. The state Supreme Court deemed his actions “reckless” and sanctioned him for contempt of court.

$5,000. Approximately 750,000 yen in Japanese currency. While painful for an individual lawyer, it is not a fatal blow.

But what if this had been a small business?

The responsibility for the lies produced by AI does not lie with the AI itself, but with the human who used it and the company to which that person belongs. Too many small businesses underestimate this fact.

—

What Happened—Sorting Out the Facts

What Aarons did was straightforward.

1. Prepared an appeal document to overturn his client’s murder conviction.
2. Asked ChatGPT to create a “perfect summary.”
3. The document generated by the AI included testimonies from non-existent witnesses and false police statements.
4. The lawyer submitted the content without fact-checking it.
5. The state Supreme Court discovered the falsehoods and imposed a $5,000 fine along with a contempt ruling.

The key point here was not “malice” but “negligence.” Aarons did not intend to lie; he simply took the AI’s output at face value.

This is what is frightening. Malice can be prevented. The thought of “I never expected AI to lie” cannot.

—

AI’s Hallucinations Are Not a “Bug” but a “Feature”

Large language models (LLMs) are designed to generate plausible text. Their job is not to produce “correct text.” Confusing the two can lead to the same situation that befell Aarons.

A similar incident occurred in New York in 2023. A lawyer had ChatGPT research case law, and it fabricated six non-existent cases. The names of judges and cited statutes were all fictional. The lawyer submitted this as is and faced sanctions.

This is not an isolated case. Anyone who has used AI has likely encountered “plausible but completely incorrect outputs” at least once. The issue arises when it is in an area outside one’s expertise—can one discern the lies?

Even lawyers could not see through it. So, can a company with ten employees and no legal department?

—

The “Lies of AI” for Small Businesses—Three Bombshells

Let’s think concretely about where these issues might explode, not abstractly.

Bombshell 1: Landmines in Contracts

Having AI draft a contract is no longer unusual. Small businesses often cannot afford to consult a lawyer for every contract, leading them to rely on AI.

But what if the contract generated by AI contained non-existent legal clauses? What if it inserted a liability waiver favorable to the other party as “standard language”?

Let’s consider the actual costs involved.

• Lawyer fees for contract disputes: Initial fees of 300,000 to 500,000 yen + success fees.
• If it escalates to litigation: 1,000,000 to 5,000,000 yen (typical for small business commercial lawsuits).
• Destroyed relationships with business partners: If a company relies on one partner for 10-30% of its revenue, that amount could vanish.

Having a lawyer review a contract costs 30,000 to 100,000 yen. If an AI-generated contract explodes, it could exceed 1,000,000 yen. In trying to reduce costs, the structure here could increase costs tenfold.

Bombshell 2: Customer Service Chatbot Run Amok

The case of Air Canada in Canada is emblematic. An AI chatbot told customers that if they purchased tickets after the death of a close relative, discounts would be applied retroactively. In reality, no such policy existed. Customers believed this response, purchased tickets, and requested discounts. Air Canada lost the case.

What “AI said on its own” is treated the same as what the company said. The significance of this being established in case law is substantial.

What would happen if a small business implemented an AI chatbot on its e-commerce site that provided non-existent return policies or incorrect product specifications?

• Costs for returns and exchanges: A few thousand to tens of thousands of yen per case.
• If it occurs collectively: Hundreds of thousands to millions of yen.
• Damage to reputation if spread on social media: Incalculable.

A chatbot costing a few thousand yen per month could lead to damages in the hundreds of thousands of yen. Because the implementation cost is low, it often runs without a checking system. This is a unique pitfall for small businesses.

Bombshell 3: Internal Documents as “Evidence”

Often overlooked, but this is the most dangerous aspect.

Internal minutes, reports, and HR evaluation comments generated by AI could potentially be submitted as evidence in labor disputes or lawsuits.

If an AI-generated HR evaluation states “performance issues persist” but contradicts actual numerical data, the credibility of the company’s claims could collapse in an unfair dismissal lawsuit.

• Settlement amounts for unfair dismissal: Even for small businesses, the market rate is 1,000,000 to 3,000,000 yen.
• Costs for labor arbitration: 500,000 to 1,000,000 yen.
• If the reliability of AI-generated documents becomes a point of contention, the cost of proving this could increase further.

Documents written by AI are considered documents written by your company. The statement “AI wrote it on its own” holds no legal weight.

—

So, What Should Be Done?—Practical Measures for Small Businesses

“Don’t use AI” is not a solution. The cost-saving benefits are real, and companies that do not use it will lose competitiveness. The issue lies in how it is used.

1. Treat AI as a Drafting Assistant

Do not make the AI output the final version. Always have a human review it. This alone could have prevented the incident involving the lawyer.

Specifically, create an operational rule that all AI-generated materials must be labeled as “draft” internally. If anything is sent out without a label, it will be treated as an incident. Simple but effective.

2. Differentiate Usage by “Explosion Radius”

• Small Explosion Radius (internal brainstorming, idea generation, rough drafts) → Fully utilize AI. Light checks are acceptable.
• Medium Explosion Radius (customer-facing emails, blog posts, proposal documents) → Use AI + mandatory human review.
• Large Explosion Radius (contracts, legal documents, HR-related matters, official responses to customers) → Use AI + mandatory expert review.

Applying the same checking system to everything is unrealistic. Adjusting for the magnitude of risk is the realistic solution for small business resources.

3. Establish Contracts and Terms Assuming “AI Will Make Mistakes”

If you are implementing a chatbot, clearly state in the terms of use that “responses from AI are for reference only and do not constitute formal contractual terms.” Air Canada lost because they did not do this.

Having a lawyer review the terms costs 50,000 to 100,000 yen. If you lose a lawsuit, it could cost hundreds of thousands of yen. Like insurance, skimping on a few thousand yen in advance can lead to hundreds of thousands of yen later.

4. Hold a “Monthly AI Lie Detection Meeting”

An elaborate auditing system is not necessary. Just once a month for 30 minutes. Pick 3 to 5 key documents generated by AI in the past month and verify their factual accuracy. Simply sharing the patterns of lies AI can produce within the company will dramatically increase literacy.

—

The Real Risk Is Not the “Fine”

A $5,000 fine is catchy as a headline, but that’s not the essence of the matter.

In addition to the fine, Aarons lost his credibility as a lawyer. His client’s appeal was ruined, directly affecting the life of a defendant who was convicted of murder.

For small businesses, “credibility” carries a different weight than it does for large corporations. Large companies can withstand scandals due to their organizational depth. If a company with ten employees is labeled as “that company that created a false contract with AI,” their business in the community could be finished.

As the cost of AI decreases, the risk of “not using it” is definitely present. However, at the same time, the risk of “using it without verification” has become equally significant.

AI is an excellent drafting assistant but a terrible decision-maker. Only companies that can make this distinction will benefit from AI.

Who is conducting the final check on AI outputs in your company? If you cannot answer immediately, that is the first task you should tackle today.