"Fairly Trained Initiative to Certify AI Models Free of Copyright Infringement"

As more and more potential legislative requirements regarding AI call for model scrutiny, some organizations are starting to offer certification programs for AI companies to demonstrate that their work does not infringe on copyright.

Bloomberg reports that Fairly Trained, founded by former Vice President of Audio at Stability AI, Ed Newton-Rex, has added a label for companies to prove that they have obtained licenses for training data protected by copyright. Newton-Rex resigned from Stability AI in November last year, claiming that generative AI "exploits creators".

The first certification, called Licensed Model Certification, will be granted to companies that use protected data to train their models. Fairly Trained claims that it will not issue certifications to developers who rely on the fair use argument to train their models.

In a blog post, Fairly Trained stated that it has already issued certifications to nine generative AI companies involved in image, music, and speech generation. These companies include Beatoven.ai, Boomy, BRIA.ai, Endel, LifeScore, Rightsify, SOMMS.AI, Soundful, and Tuney.

Training AI models using data protected by copyright has been a challenging issue in the field of generative AI, especially after artists and authors sued several AI companies for copyright infringement. The New York Times also sued OpenAI and Microsoft for infringing its copyright while training their GPT model.

Some proposed bills currently being pushed in the US Congress will require AI companies to disclose the sources of their training data. This information will allow copyright holders to know if their works are being used without permission.