An overview of the legal framework for generative AI

How to balance generative AI and respect for the law?

In the absence of specific legislation on the use of artificial intelligence, common-sense rules applicable to any creative work apply. Let's try a quick test:

If you are producing content for a client:

• Make sure that your creations don't already exist elsewhere?
• Are you clear on the methods of transferring the rights to your creations?
• Are you aware of the risks in case of counterfeiting, or if your client is sued?
• Do you have professional liability insurance?
• If you use service providers: do you carefully review their terms and conditions of sale?

If you answered yes to all these questions, then good news, you already know how to work with generative AI!

Traditional intellectual property rules apply

Simply put, all the constraints you have in mind during your "traditional" creative processes are the same. Consider artificial intelligence as a service provider, and apply the same common-sense rules.

A brief summary of key points to be aware of:

Prior art: ensure that your creations do not already exist.
Assignment of rights: research the methods for assigning rights to your creations.
Client contract: pay close attention, particularly to risk clauses.
Insurance: professional liability insurance is, of course, essential.
Service providers: carefully review contracts with your service providers.

A new rule is added to these traditional rules of intellectual property, in order to adapt to new uses integrating AI: documentation.

Documentation, an essential reflex

As we know, the field of AI-powered creation is still undergoing a major transformation. And the law, as usual, will adapt to its use. Claire Caquant's most sensible advice is this: document all your AI creations. In practice, this means archiving all your prompts and creative processes using AI:
Keep all your prompts in a documented file, and if necessary, take screenshots of your conversations with the AI ​​to gather evidence of your creative process. This evidence can protect you in the event of legal action.

Let's take a simple example: it's probably very risky, in terms of intellectual property rights, to ask a generative AI to create an image of Hermione Granger, a registered trademark. On the other hand, you can ask the AI ​​to create an image of a young schoolgirl with curly red hair and freckles, wearing a British school uniform with its red badge on the front, holding a wand.

The image will likely be essentially the same, but the legal risk becomes nil. It will be up to you to prove that your prompts were within the bounds of intellectual property.

What is the current legal framework for generative AI?

Let's take a step back and recall the three levels of legal protection in terms of intellectual property, and its applications to artificial intelligence.

Three legal pillars must be respected in intellectual property law:

The protection of fundamental rights

This translates into the AI ​​Act, adopted by Europe in July 2024. It is the first regulation in the world that guarantees the protection of fundamental human rights, safeguarding democracy, health, and the environment. Among other things, it prohibits biometric categorizations based on political, religious, or philosophical opinions, sexual orientation, or ethnicity; citizen rating systems; and mass surveillance for creating databases of facial recognition, emotion recognition, and other technologies.

The protection of fundamental rights

It relates to intellectual property: how I protect my creations, how third parties protect theirs, and how I protect myself if my data is hacked. Legal battles are currently underway in this area of ​​law, particularly in the United States. In summary:

• Language model learning currently operates like a black box. Model training itself isn't illegal, but this is precisely where the lawsuits are focused. The legal loophole is very narrow since the training is done using web crawlers, something search engines like Google do constantly to index the web, without this activity being considered illegal.
• However, content generation must respect copyright and is not protected by copyright.
• If the generated content is retouched or modified , the work becomes protectable because there is human work involved.

The GDPR

The General Data Protection Regulation (GDPR) governs data processing in Europe. In this respect, things get a bit more complex since many language models are outside of Europe. The key question to keep in mind is where your data goes as soon as you send a prompt to an LLM server!

• Do your actions train the model? If you're using the free version of ChatGPT, you train the model with every action. With the paid version, you have to uncheck the option via a somewhat hidden opt-out. For Mistral, you need to send an email. Each service has its own system; you'll need to find the answer to protect yourself.
• Is the LLM server located in the European Union? This question becomes crucial in the case of sensitive data, as the EU has more protective guarantees for rights holders.

In summary, to be “GDPR Compliant”, the actions must not affect the model and the servers of the model must be located in the European Union.