AI companies to disclose copyright training data in United States

Copyright infringement concerns: Disclosing US Representatives Anna Eshoo and Don Beyer have introduced the AI Foundation Model Transparency Act to enhance transparency and address concerns over copyright infringement. This legislation would compel creators of foundational AI models to disclose the sources of their training data.

AI Foundation Model 5 Transparency Act of 2023 | Image credit: Beyer.house.gov

Under the proposed act, companies developing these AI models will be obligated to disclose details such as the origin of training data, data retention protocols during the inference process, model limitations or risks, alignment with the National Institute of Standards and Technology's (NIST) AI Risk Management Framework, and any other federal standards that may be instituted. Companies must also specify the computational resources used for training and operational phases.

The bill also underlines safeguarding against misinformation, requiring AI developers to conduct rigorous testing, or "red teaming," particularly in sensitive areas such as healthcare, cybersecurity, elections, and areas affecting vulnerable populations like children.

Copyright disputes

Recently Business and Property Courts of England and Wales have decided to delve deeper into Getty's allegations against the startup regarding the use of its copyrighted image library to train the Stable Diffusion model. Stability contended that the matter should not be adjudicated in a UK court, asserting that none of the individuals involved in developing or training the Stable Diffusion model were located in the UK. 

On the other hand, a judge in California's federal court has shut down a lawsuit by digital visual artists who accused Stability AI, Midjourney, and DeviantArt of using their copyright work to train their models and generate AI images. 

The judge said the artists could file an amended complaint against the two companies, whose systems use Stability's Stable Diffusion text-to-image technology.

Highlighting the context for such legislative action, the bill references recent copyright infringement lawsuits against AI companies. The bill asserts, 

"Public use of foundation models has led to countless instances of the public being presented with inaccurate, imprecise, or biased information.”

The legislation will complement the Biden administration's executive order on AI, which aims to establish reporting standards for AI models but needs the enforceability of law. The AI Foundation Model Transparency Act now awaits assignment to a committee for further deliberation, with its timeline potentially affected by upcoming election campaign activities.