What does “general-purpose AI” mean?
General-purpose AI (GPAI) describes AI — typically large models trained on broad data at scale — that displays significant generality: it can competently perform many distinct tasks (drafting, coding, analysis, image generation) rather than a single narrow function, and can be integrated downstream into countless applications. The term entered EU law to separate the model layer from the application layer: the same foundation model may power a harmless writing aid and a high-risk hiring tool.
Concept versus statutory definition
This entry covers the concept. The operative legal definition and the obligations — Article 53 documentation, copyright policy, the training content summary and the systemic-risk tier — are covered in our canonical entry, GPAI model (EU AI Act definition). The conceptual point that survives into the statute is generality itself: the law attaches duties at the model layer precisely because nobody can enumerate a general model’s uses in advance.
The legal dimension
The layer separation does real work. Model-layer duties — documentation, copyright, transparency towards downstream builders — sit with the model’s provider and are supervised centrally; application-layer duties follow the risk classification of each system built on top. A builder therefore reasons in two steps: what does my model’s provider owe me, and what does my application’s risk class require of me? Neighbouring concepts mark out the map: the foundation model is the technical substrate, the large language model the dominant GPAI family, and AI agents are GPAI given tools and autonomy.
Turkish context
Türkiye has no GPAI regime of its own — no AI-specific law is in force — so the concept matters to Turkish teams through the EU: a product sold into the Union inherits the layer logic, and a Turkish company that substantially modifies a general model can find itself holding model-layer duties. Domestically, KVKK and general provisions govern what the model does with personal data.
Do: keep the two layers separate in your compliance file — model questions go to the vendor, application questions to your own risk class. Don’t: treat “GPAI” as a marketing synonym for “AI”: in EU law it is a defined regulatory category with its own supervisor.
Related guides: You Didn’t Train the Model, but You Still Have AI Act Obl….
Working on this? Vircon Legal advises on AI & Algorithm Law and AI Compliance Hub. Talk to us →
If this is on your desk
Templates and checklists are free in the Founder Academy; for a specific situation, book a 30-minute intro call.
Founder AcademyBook an intro call