Glossary
Updated
Under the EU AI Act, General-Purpose AI (GPAI) models, like LLMs, face specific transparency and copyright obligations. If a model reaches the Systemic Risk threshold (e.g., 10^25 FLOPs), it requires additional red-teaming and safety testing. Most enterprise agents today are built on top of third-party GPAI models, making this category centrally important to enterprise AI governance.
Example
OpenAI's GPT-4 is a GPAI model; when a company uses it to build a specific Customer Support Bot, the company must manage both the GPAI provider risk and the specific application risk.