Skip to main content

AI models that don’t violate copyright are getting a new certification label

AI models that don’t violate copyright are getting a new certification label

/

Nonprofit group Fairly Trained plans to certify AI models that ask permission to use copyrighted material.

Share this story

AI brains in a network
Illustration by Alex Castro / The Verge

As much of the potential legislation around AI requires reviews of models, groups have begun offering certification programs to AI companies to show how their work doesn’t violate copyright.

Bloomberg reported Fairly Trained — founded by former Stability AI vice president for audio Ed Newton-Rex — adds a label to companies that prove they asked for permission to use copyrighted training data. Newton-Rex started Fairly Trained after he quit Stability AI in November, citing generative AI was “exploiting creators.” 

Its first accreditation, which it calls the Licensed Model certification, will be awarded to companies that license protected data to train its models. Fairly Trained claims it will not issue the certification to developers that rely on the fair use argument to train models. 

Fairly Trained said in a blog post that it has already certified nine generative AI companies that work in image, music, and voice generation. These include Beatoven.ai, Boomy, BRIA.ai, Endel, LifeScore, Rightsify, SOMMS.AI, Soundful, and Tuney. 

Training AI models with copyrighted data has been a fraught issue in generative AI, especially after artists and authors sued several AI companies for copyright infringement. The New York Times also filed a lawsuit against OpenAI and Microsoft for violating its copyright when training its GPT models. 

Some proposed legislation currently making its way through Congress would require AI companies to disclose where they got their training data. The information will let copyright holders know if their work was used without permission.