The Computer & Communications Industry Association (CCIA) has launched a new global template aimed at enhancing transparency in the development of general-purpose artificial intelligence (GPAI) models. This initiative is designed to increase trust and improve understanding of AI training data.

The newly introduced Template for Transparency in AI Model Training Data will require companies to disclose the types of data used for training GPAI models, ranging from podcasts to books. Companies will also need to reveal the main data sources, such as open-access archives or large public datasets, and explain how this data is selected and employed for model training.

This transparency template is based on eight core principles endorsed by CCIA’s members, which include leading AI developers. One key principle emphasizes that "transparency measures may not compromise AI model safety, trade secrets, or confidential business information of both model providers and third parties."

The CCIA initiative aims to contribute to the European Commission's ongoing work on a similar template as part of the AI Act’s implementation. The association stresses that it is important for the EU to align with international approaches rather than setting standards independently.

Boniface de Champris, CCIA Europe’s Senior Policy Manager, stated: “CCIA wants to foster the responsible public disclosure of AI training data. Our transparency template provides a concise and effective way to report on training data while safeguarding the security of AI models, companies’ valuable trade secrets and confidential information.”

Joshua Landau, CCIA Senior Counsel for Innovation Policy, added: “Transparency measures can be helpful in maximizing the potential of AI, but these measures must be technically reasonable and proportionate. Striking the right balance is crucial to build trust and ensure compliance with legal and ethical standards while still encouraging innovation in artificial intelligence.”