General-purpose AI model and its obligations

A general-purpose AI model is a model capable of performing a wide range of tasks that serves as the foundation for a large number of specific applications. European regulation treats it as a separate category, because its properties carry through into everything built on top of it. The obligations are directed primarily at the model provider – they concern mainly technical documentation, information for those who integrate the model, compliance with copyright in the training data, and publication of a summary of the data used, with stricter requirements added for models with systemic risk. For an ordinary company, the practical consequence that matters is that documentation from the model provider is an input into your own documentation, and when choosing a supplier it is worth checking whether they make it available at all.

See also: AI Supply Chain, Technical documentation of an AI system, AI risk classification.