top of page
Ravit banner.jpg

RESOURCES

Groundbreaking fine for training on copyrighted data

Huge news! French regulator fines Google €250M for training on copyrighted data. It’s the first genAI copyright fine I know of.




➤What happened:


👉 France’s competition authority found that “Google trained its Bard AI on copyright news articles without giving publishers sufficient information about remuneration or an opportunity to opt out”


👉The fine comes after Google failed to comply with a settlement in 2022. Back then, Google avoided this very same fine by pledging to enter good-faith negotiations with news providers about compensation for training on their content and other issues. Google failed to do so. 


👉Therefore, the French competition authority fined Google for: “using news content to train its Bard AI service (now called Gemini), without the permission of the publishers, and without providing access to an opt-out tool that would have let them contest the AI usage.”



➤Why this is huge:


📣When it comes to copyright law and generative AI, most of the activities so far are many lawsuits, especially in the US, and often skepticism about whether and how copyright laws apply to AI.


📣This is the first time (that I've heard of) that a company has been penalized for the unauthorized use of data for training.


📣The fine was announced in March, but I only heard about it now (by accident!), which leads me to believe that some underappreciate its significance. 


📣This fine is a potential game-changer! 


Will other countries follow suit? Will this fine impact the many genAI copyright lawsuits? Will it make companies improve their training practices? Time will tell!



➤Read the full story here



➤ *** Two updates ***


1. There was one other genAI copyright fine, in China, February 2024 - In this case, a company sued because genAI images were too similar to their copyrighted images. The judge awarded the plaintiffs $1,389 in damages. Note that here the genAI company was just required to have a mechanism to stop spitting out certain images, but with no restrictions on the training. Thank you to Jaya Files for sharing this with me!


2. In response to some comments - This fine is an important precedence because it is the first case (that I know of) where regulators determined that training on copyright-protected data is illegal. Even though €250M is not a lot for a company like Google, the penalties can accumulate if others fine them on similar grounds or if this fine is used to support court cases globally. In addition, it is interesting that it is the agency that fined them is a competition authority. There might be ways to forbid training on copyrighted data regardless of copyright laws, using other laws such as fair competition laws (I couldn't figure out which law exactly the agency used, please share if you know!)



➤Insight and reflections welcome! Join the conversation in the Linkedin thread!


Comments


FOR UPDATES

Join my newsletter for tech ethics resources.

I will never use your email for anything else.

bottom of page