Authors' coalition scorns EU rules that tore up AI protections five years ago today
Copyright directive 'paved way for the plunder of cultural works'
WHAT’S HAPPENED?
A BROAD COALITION of authors’ groups has launched a blistering attack on Europe’s copyright directive which came into force five years ago today, saying it hands AI giants a “licence to steal”. The European Writers’ Council (EWC), a federation of 52 associations and unions from across the European continent, is now calling on the European Commission to strike references to the directive in the EU AI Act and enable “entirely voluntary licensing”.
Agreed in 2019 — long before the arrival of the AI boom — the EU copyright directive introduced an exception for text and data mining (TDM) giving creators a theoretical opt-out. In 2024 the EU AI Act then referred to the directive. Last year, Axel Voss, the German member of the European Parliament who helped write the directive, said it was only intended to have limited uses and shouldn’t have been seen as being part of the bloc’s flagship AI legislation (Charting #42).
The EWC, whose members represent more than 250,000 authors across all genres, said the “flawed over-interpretation” had “paved the way for an all-you-can-read free buffet” for non-European tech firms. EWC president Sebastià Portell, an award-winning writer and translator, said the Commission had failed to accept that the TDM exception shouldn’t apply to AI development.
“Authors are plundered on a historic scale and their right to voluntarily opt-in under fair licensing conditions is ignored,” he said, adding that the “non-authorised, non-remunerated and utterly opaque exploitation of human endeavours” amounted to a “transfer of value towards foreign tech”.
Whatever benefits the directive may have brought in a few individual cases, this ‘licence to steal’ completely nullifies all good intentions.”
Nina George, the international best-selling author and political affairs commissioner at the EWC, told Charting the continued misinterpretation of the TDM exception in the EU AI Act was “no surprise”.
“If you build a future on a lie, you have to continue pretending it’s true, otherwise your world implodes and you lose face. If the Commission finally had the dignity to say ‘Oops, pardon, my bad’, then all existing generative technologies would need to be shut down to not risk claims on copyright infringement running to trillions of Euros.”
George, who heads the EWC’s AI task force, said the “crisis” over transparency “keeps on going”. While the directive gave rightsholders an opt-out, a lack of transparency meant it was impossible for them to know if their works had been used for AI training. Last year’s transparency template, introduced by the Commission, only forces AI developers to list the top 10% of domain names (but not individual book titles) that their crawlers have scraped. AI providers who can’t provide training details are allowed to just state they don’t have the information, or that providing it is too much of a burden (Charting #64).
In February 2025, Sebastian Stober, artificial intelligence professor at the Otto von Guericke University Magdeburg, established it was both technically feasible and “in many cases” a “trivial” task for AI developers to provide a sufficiently detailed list of works used to train their models.
“Poor things, they might have to use — hey! — AI,” quipped George who contrasted the “relaxed” treatment of AI developers with “costly opt-outs on every title and every format” imposed on creators who still lacked a standard and robust means of asserting their rights that wasn’t then ignored by AI firms.
“As long as there is no confirmation by AI developers to having acknowledged the opt-out, there is neither a way to licensing, nor to examine one’s right to not be eaten up by a product spitting out competing and plagiarising products in minutes.”
Last month the Commission launched a consultation on the EU’s copyright framework following an overwhelming vote in the European Parliament backing a report authored by Voss. The potential breakthrough report called on the bloc’s civil servants to find a “permanent” solution to ongoing copyright exploitation.
The EWC wants the Commission to remove the “non-applicable TDM exception” in order “to allow entirely voluntary licensing by authors” — in other words, an opt-in. It is also calling on commissioners to beef up transparency requirements, and remind cultural and heritage institutions such as libraries, archives and museums that they are not permitted to make rightsholders’ copyrighted works available to AI developers without the prior consent of authors.
And what if the Commission ignores creatives? George told Charting:
“We take the AI companies to court, one by one. Burn millions of Euros, get our dignity back, while never actually being paid back. But hey, we need to fight for all authors and artists that follow after us, and prefer to do the good things in life, like thinking, expressing, creating, and making the worthwhile mental and emotional connections between humans.”
WHY SHOULD WE CARE?
✨To think there was a time, not so long ago, when Europe’s approach to creative copyright in the era of generative AI was regarded as a template for others to follow. The UK’s recently abandoned TDM exception with opt-out bore an uncanny resemblance to the EU copyright directive’s reserved rights model. The UK’s AI Opportunities Action Plan in January 2025 urged the government to follow the EU’s lead, saying its TDM exception was “designed to support AI innovation while also enabling rightsholders to have control over the use of content they produce”. Absolute piffle, of course, and another sign that those advising ministers were parroting the lobbying mantras of Big Tech. George told Charting that the EWC will now be compiling red lines that the Commission would be “well advised not to cross”, such as a new AI exception, or collective mandatory licences for national large language models (LLMs) as member states seek to lessen their dependence on Silicon Valley and develop sovereign AI approaches. Both of those concerns apply beyond Europe. While the UK has dropped its much-derided TDM exception with an opt-out that would have been impossible to manage, it hasn’t yet said what will replace its once-preferred option, nor fully ruled out a new exception. And creators need to watch any attempt to introduce mandatory licensing in the name of ‘innovation’ like hawks.
RELATED:
▪️Charting Gen AI’s Weekly Newsletter on Friday will have full coverage of the week’s developments. Subscribe now so you don’t miss it!
In case anyone doubts what I wrote above, here’s the link to the IRS nonprofit entity search site. Search by company name for OpenAI.
I’m of half a mind to file a FOIA (freedom of information act) request for their full and complete application, as well as for the pages missing from their approval letter from the IRS.
https://apps.irs.gov/app/eos/details/
The original lie, at least on this side of the Atlantic, was that OpenAI was a 501(c )(3) nonprofit established for the purposes of research only. That it would not engage in for-profit activity. And that its use of copyrighted material was for research only and would not be used in a for-profit business.
Failure to challenge that lie has resulted in everything we see now.