AI-generated texts pose legal risks that companies must manage

You know that feeling of relief when a text seems to write itself? You ask the AI, get a paragraph that seems perfect, and think it’s exactly what you needed. Then, a few days later, the warning comes: it contains a regulation that doesn’t exist or a sentence copied from a competitor’s website. That sense of relief quickly turns into a real problem. Using artificial intelligence to write can give you an edge, but only if you know how to check the content before using it. Otherwise, the legal risk becomes very real.

What Are AI “Hallucinations” and Why Can They Harm You?

Hallucinations are responses that AI presents as correct and authoritative, but which are actually false or completely fabricated. This happens when the model doesn’t have a definite answer and constructs information that seems plausible, without warning that it’s creating it out of thin air. As reported by Brocardi.it, the risk is very real: in a case examined in Ferrara, generative AI software had cited a ruling by the Venice Court of Appeals with a specific case number and date. That ruling, however, had never been issued. The lawyer who used it without verifying the information took a huge risk. If this can happen in a legal document, it’s easy to imagine what happens when the same thing appears in a company communication or in a contract you send to a client. Never assume that what AI produces is 100 percent accurate.

The Risk of Plagiarism: It’s Not Just an Academic Issue

Many people associate plagiarism only with college theses, but for a company, the problem is much more immediate. When you ask AI to write a text and it reworks others’ sources without citing them, you may end up infringing on copyright. As explained by Tesify, the proper use of AI should be similar to that of a tutor: it helps you organize your thoughts, but the responsibility for the final result remains yours. The problem arises when you ask AI to write an entire chapter, to generate notes without oversight, to automatically rewrite others’ texts to hide their source, or to create bibliographies that you haven’t verified. If your website or brochures contain copied text, you risk penalties and reputational damage that’s difficult to repair. Unintentional plagiarism is even more insidious because you don’t do it on purpose, yet the legal consequences remain the same.

How to Verify AI-Generated Text in 3 Steps

You don’t need to be a legal expert—you just need a method. Before publishing or sending any document, you can follow these three steps. Start by verifying the sources. If the AI cites a law, an article, or a court ruling, look up the original reference in official databases or on institutional websites. Never trust the link provided by the AI, because it often appears correct but leads nowhere. Next, cross-check the information. Compare what you’ve found with at least two different reliable sources. If a piece of information appears only in the AI’s response and nowhere else, that’s a red flag you shouldn’t ignore. Finally, you can use plagiarism-checking software. These tools help you determine whether the text is too similar to content already available online. They aren’t infallible, but they offer a useful initial indication.

Transparency: The New European Code of Conduct

Europe has introduced new rules for labeling AI-generated content. According to the Code of Conduct on Transparency, companies must be transparent when using text created by automated systems. The code is divided into two parts: one for AI system providers and one for users. For business owners, the key point is simple: if text published on your website or sent to customers was generated by AI, you must be able to disclose that fact. This isn’t a requirement that puts you at a disadvantage—quite the opposite. This transparency builds customer trust, as customers value honesty. These rules help people recognize when content has been created or modified by AI, reducing the risk of deception and protecting you from future disputes.

How to Manage Customer Service Without Falling Prey to “Hallucinations”

If you use AI to respond to customers, there’s always a risk of providing incorrect information. A customer who asks about the return policy and receives a procedure invented by the AI will immediately turn an automated response into a complaint. To avoid these issues, the most effective solution is to integrate tools that allow for constant human oversight. If you want to streamline conversation management without having to manually verify every response, platforms like Leader24 let you manage the handoff between AI and a human agent: when the AI is unsure, the system assigns the conversation to you, so the customer always receives verified responses. The principle remains the same: the AI does an initial screening, but the final responsibility is yours. This approach protects you from legal risks.

The First Step to Take Today

You don’t have to stop using AI, but you can change how you use it. Start treating every response as a draft to be reviewed, not as text ready for publication. Create a simple checklist that everyone on your team follows before sending an automatically generated text. Three checks are enough: verify dates and legal references, check citations and sources, and ensure the tone and consistency align with your usual communication style. You don’t need a legal advisor for every email. All you need is a method and a little extra attention before clicking “send.”

Frequently Asked Questions

Can AI really invent a law or a court ruling?

Yes. The phenomenon of “hallucinations” has been documented even in real-world contexts. According to 24 Ore, an AI program cited a Venice Court of Appeals ruling in a legal document, complete with a specific case number and date. That ruling did not exist. If it can happen in a courtroom, it can also happen in your business communications.

How can I tell if text generated by AI is plagiarism?

You can use a plagiarism-checking tool. It won’t give you absolute certainty, but it will flag suspicious similarities with content already published online. The critical threshold is generally set below a significant similarity percentage: below that percentage, you’re in the clear; above a significant threshold, you run risks. More important than the number, however, is the type of content: a sentence taken from a competitor’s website is more dangerous than a generic phrase.

Do I have to disclose that a text was generated by AI?

The new European code of conduct on the transparency of AI-generated content recommends it. It’s not yet a legal requirement in all cases, but it remains a best practice that protects you from disputes and strengthens customer trust. When the text is intended for the public, disclosing its origin sends a message of honesty that your customers appreciate.

Leader24 Insights

If you’d like to learn more about how Leader24 addresses these topics, here are some resources to get you started:

• The sectors in which we operate
• Transparent pricing