The EU Plans to Force AI Image Generators to Disclose Use of Copyrighted Content in AI Training

We are getting closer and closer to having a defined legal frame for AI-generated visuals, as the EU recently added a rule to their EU AI Act that forces all AI tool developers to disclose the datasets used to train their software – including any copyrighted work. 

If this law passes, it would mean that apps like Midjourney and Dall-E, which so far have not made public any details on their datasets, will be obligated to share this information and –undoubtedly, at this point– disclose millions of unauthorized uses of copyrighted images. 

Let’s have an in-depth look at this. 

The EU AI Act – A Proposal for Regulating AI Technologies

The European Union’s Artificial Intelligence Act (EU AI Act for short) is a law that has been in the making for two years and is set to be introduced this spring. 

This is said to be the first attempt by a major law-regulating institution at framing the novel AI technology from a legal standpoint. European legislators have shown prevision of the future by starting drafting it when this kind of tech was still in its infancy. 

The proposed regulation will evaluate AI tools on relevant factors like biometric surveillance, spreading misinformation, and discriminatory language and rate them based on their perceived risk – minimal, limited, high, and unacceptable. However, no tool would necessarily be banned: high-risk applications would only be required additional transparency. 

Per this proposal, AI apps that wish to operate in the EU must comply with these regulations or otherwise pay hefty fines (according to MIT Technology Review, fines could be up to 6% of the developer’s total annual turnover).

The AI Act Will Now Require Disclosing AI Training Datasets 

The latest draft of the EU AI Act, which was worked on for the past couple of weeks and revealed last Friday, has added a point that requires all AI generative model developers to disclose their data pools and any copyrighted work included in them. 

As reported by Reuters, this requirement reaches both AI text and image generators, but it seems to have a greater impact on the latter, as image copyright is a much more straightforward matter to rule on. 

As you probably know, some of the most popular and advanced AI image generators available to the public have never fully disclosed which content they used to train their models, rather have given vague answers: Midjourney’s CEO, for example, only ever said that they used “hundreds of millions of images” from the web… without permission. This new disposition from the act would force them and other user-favorite apps like OpenAI’s Dall-E to say exactly which content they used in their training process. 

It’s worth noting here that this regulation isn’t banning the use of copyrighted material in AI training, merely forcing said use to be made public knowledge. 

What Could This Disposition Mean for AI Image-Generating Tools

As there is an unsaid truth regarding AI image generator developers pooling copyrighted images without authorization or compensation into their datasets —Midjourney has recently made headlines for fake images of famous people made with their app, which would be impossible without training the model with copyrighted images of those people–, such a law is expected to open Pandora's box of copyright lawsuits against AI apps in the EU. 

As we reported earlier this year, Getty Images has already filed lawsuits in the UK and the US against Stability AI's Stable Diffusion, precisely for using the agency’s copyrighted visual content in their training datasets without permission. There is little doubt that if they (and other companies and artists) find their protected work in use at other apps, they’d also take legal action against them. 

On the other side of the coin, though, this could also give AI generative model developers more incentive to resort exclusively to authorized content for their dataset needs, something other companies like Shutterstock or Adobe are already doing. 

What are your thoughts on the latest EU AI Act’s dispositions? Share them below!

THE AUTHOR

Ivanna Attie

All About Ivanna

I am an experienced author with expertise in digital communication, stock media, design, and creative tools. I have closely followed and reported on AI developments in this field since its early days. I have gained valuable industry insight through my work with leading digital media professionals since 2014.

• NETWORKING