The EU’s AI Act has reportedly added a new addition to the legislation that could have a large impact on AI image generators regarding copyright. The late addition would allegedly force companies such as OpenAI to disclose their use of copyrighted training data.
Obviously, this could have a huge impact and start a chain reaction of copyright lawsuits against the image-generating companies. Large stock image companies such as Getty have already started the process of suing Stable Diffusion over their use of copyrighted images to train their AI.
So far, the majority of AI text-to-image companies have freely admitted to scraping data from copyrighted images online. The only one to stand out against this trend is Adobe’s Firefly, which the company says was only trained on Adobe’s own licensed stock and non-copyrighted images.
This unauthorised use of intellectual data and images has sparked a fierce debate across the art world. Up until now, OpenAI has avoided scrutiny by simply refusing to disclose its training data. If this new EU legislation is passed, that could change.
According to Reuters, the AI Act was approved in draft earlier this week and will require “companies deploying generative Al tools, such as ChatGPT … to disclose any copyrighted material used to develop their systems.”
The AI Act has been in the works for the past two years, although due to the extremely rapid advancement in 2023, it has been pushed to the top of the list. The Act will cover all aspects of AI and will classify each technology under a perceived level of risk in terms of biometric surveillance, spreading misinformation or discriminatory language.
Rather than outright banning AI generators that have used copyrighted material, the EU is asking for transparency. However, it is unlikely that companies will be required to comply with the new laws until at least 2025. By then, we could be looking at a very different AI landscape from now. The question is: will this be too little, too late?