The European Union (EU) bloc’s Artificial Intelligence (AI) Act is leaving behind a trail of devastating copyright gaps, according to one of the authors behind the legislation. The piece of regulation was intended to protect writers, creatives, musicians, and all artists, something that is not the case now that the law is in place, due to an “irresponsible” gap.
The EU law was not conceived to deal with generative AI. According to The Guardian, 15 cultural organizations have penned a letter to the EU this week expressing their concerns, warning the bloc that the draft rules to implement the AI Act were “taking several steps backward” on copyright. Another writer spoke of the “devastating” gap.
A German center-right member of the European Parliament, Axel Voss, who was crucial in the writing of the EU’s 2019 copyright directive, revealed that the legislation was not perceived to tackle generative AI models. Voss revealed that the “legal gap” came to light after finalizing the EU’s AI Act. This, according to Voss, meant that copyright was not enforceable in this area.
The EU’s AI Act was put into effect in 2024. The law was, however, already in the making when OpenAI’s ChatGPT launched in November 2022, to immediate success, becoming the fastest-growing consumer application in history. OpenAI is also the creator of the image generator Dall-E.
The AI model, capable of generating images, text, and job applications, together with other AI systems, was trained on vast troves of books, newspaper articles, images, and songs. The rise of these models has also caused alarm and disgruntlement among newspapers, authors, and musicians, triggering a chain of lawsuits about alleged breaches of copyright.
It’s hard to believe the legal gap remained. According to Voss, he was unable to get the majority of EU lawmakers to ensure strong copyright protection when the matter arose in the late stages of negotiating the AI Act. He revealed that the lack of strong provisions on copyright was “irresponsible” and it was “unbelievable” that the legal gap remained.
According to The Guardian, Voss would like to see the gap filled, although it would take years following the EU’s decision to withdraw the proposed AI Liability Act. According to the AI Act, tech firms must comply with the 2019 copyright law, which includes an exemption for text and data mining (TDM). Voss pointed out that this exemption from copyright law was meant to have limited private use, rather than allow the world’s largest firms to use vast amounts of intellectual property. He added the introduction of the TDM exemption in the AI Act was “a misunderstanding.”
An academic study done last year by legal scholar Tim Dornis and computer scientist Sebastian Stober reinforces this as it concluded that the training of generative AI models on published materials could not be considered “a case of text and data mining” but “copyright infringement.”
A German bestselling author and president of honor at the European Writers Council, Nina George, whose creations have been translated into 37 languages, described the TDM exemption as “devastating.” “These AI exceptions for commercial use mean that business interest will be served for the first time,” she said, adding the exclusions from copyright were initially meant to balance the interests of artists against those of the public, for example allowing schools to photocopy texts. She added she had no way of ascertaining if any of her works had been used to train generative AI systems.
“The lack of instruments to enforce any rights, this is the scandal in the construction of the AI Act [in] relation to copyright directive,” she added. This comes as companies are not under obligation to disclose the content they use to feed their generative AI models. According to The Guardian, starting from August 2, tech firms will have to give a summary of data used in AI models, although details are still being decided.