Stay updated on AI and copyright through our newsletter (danish).
Our insights into how AI System Owners use illegal content to train their AI models have been met with strong demand. In a new video presentation, we share our key findings on this challenge.
Many of the world’s leading generative AI System Owners, including Meta, OpenAI, Anthropic, and Midjourney, have scoured the internet, including and intentionally the obviously illegal parts, so they could build massive datasets with copyrighted content to train their AI models. All done without permission and payment to rights holders.
This has been documented by the Danish Rights Alliance in several reports, which has generated significant interest and demand for insights into our work from stakeholders in Denmark and abroad.
We are therefore releasing a video presentation that conveys the insights which the Danish Rights Alliance’s Head of Content Protection and Enforcement, Thomas Heldrup, has been presenting over the past years to several rights holder organizations and policymakers around europe.
Watch the full presentation in the video here:
In the presentation video, Head of Content Protection & Enforcement, Thomas Heldrup, highlights our key findings on how AI System Owners use copyright-protected content, sourced from illegal file-sharing services, to train their AI models.
In the report on pirated content used in the training of generative AI, you can explore our research in more detail.
You can also read about our key focus areas in this field, and how we work to safeguard copyright in relation to artificial intelligence, here.