General-purpose AI (GPAI) / foundation models
General-purpose AI (GPAI), also called a foundation model, is an AI model trained on broad data that can be adapted to many downstream tasks — and under the EU AI Act, its providers carry specific transparency and documentation obligations.
GPAI models — the large language and multimodal models behind many modern AI features — are not tied to a single purpose, so the EU AI Act regulates them separately from application-level systems. Providers must supply technical documentation, information for downstream developers, a policy to respect EU copyright law, and a summary of training data.
Models judged to pose systemic risk (based on capability thresholds) face additional duties: model evaluation, adversarial testing, incident tracking, and cybersecurity safeguards.
The obligations flow down a supply chain: if you build on someone else's GPAI, you inherit responsibilities as a deployer even if you did not train the model. See the EU regulatory framework for GPAI-specific rules.
