General-Purpose AI

What is General-Purpose AI?

General-Purpose AI (GPAI) is a category specifically defined by the AI Act for AI models that can be used for a wide range of different tasks. Language models typically fall into this category.

How the obligations are structured

General-Purpose AI models are subject to baseline obligations such as technical documentation and information obligations toward downstream providers. Models classified as General-Purpose AI with systemic risk—among other things based on the amount of computing power used during training—are subject to additional, more stringent requirements.

Practical significance

A Foundation Model describes a technical state of a model—the stage immediately after pretraining, before Fine-Tuning. General-Purpose AI, by contrast, describes a legal category under the AI Act based on a model’s general capabilities, broad range of applications, and risk criteria. Both terms can apply to the same system, but they describe different characteristics. A provider of a General-Purpose AI model has its own legal obligations, regardless of who later deploys the model.

Example

A large language model can technically be a Foundation Model while simultaneously being classified as General-Purpose AI under the AI Act. The two terms describe different characteristics of the same system rather than competing definitions.

Further reading

Related terms

→ AI Act
→ Provider
→ Foundation Model