The bill emphasizes the need for transparency in training data due to increasing copyright infringement lawsuits against AI companies. It cites cases involving Stability AI, Midjourney, Deviant Art, and Getty Images. The legislation, which still needs to be assigned to a committee for discussion, complements the Biden administration’s AI executive order on reporting standards for AI models. If passed, it will make transparency requirements for training data a federal rule.
Key takeaways:
- Reps. Anna Eshoo and Don Beyer have filed the AI Foundation Model Transparency Act, which would require creators of foundation models to disclose sources of training data to protect copyright holders.
- The bill would require companies to report on how data is retained, the limitations or risks of the model, alignment with NIST’s planned AI Risk Management Framework, and the computational power used to train and run the model.
- The bill emphasizes the importance of training data transparency in relation to copyright, citing several lawsuits against AI companies for alleged copyright infringement.
- The bill complements the Biden administration’s AI executive order, but if passed, it would make transparency requirements for training data a federal rule.