GPAI model with systemic risk
A GPAI model with systemic risk is a general-purpose AI model that may have a significant impact on public health, safety, public security or fundamental rights, due to its scale. This is presumed where the cumulative computation used for its training exceeds 10^25 floating point operations; in that case the provider is subject to the four additional obligations of Article 55.
The concept and threshold are set out in Article 51(2), the rebuttal procedure in Article 52, and the obligations in Article 55 and Annex XIII. The 10^25 FLOP threshold refers to the cumulative amount of computation used to train the model — a scale currently associated with the largest, most publicly discussed language models. The presumption is rebuttable: the provider may argue, with substantiated reasoning, that the model does not in fact pose a systemic risk, though the Commission may reject that argument. Where the threshold is met, or the Commission designates the risk by individual decision, the provider must notify the Commission within two weeks and comply with the four additional obligations of Article 55: model evaluation to identify and mitigate systemic risks, assessing and mitigating those risks, tracking and reporting serious incidents, and ensuring an adequate level of cybersecurity protection. The GPAI Code of Practice offers a voluntary, but practically accepted, compliance pathway for these duties.
This category does not directly affect Hungarian SMEs in practice — it only becomes relevant if the company itself develops or trains a large-scale model, which is an exceedingly rare case.
Need documented AI-literacy training?
Article 4 is a duty of diligence: what counts is not knowledge in the abstract, but demonstrable, documented effort. Our starter package lets you begin free.
Start freeRelated terms
- General-purpose AI (GPAI)A general-purpose AI model (GPAI) is a model that displays significant generality, is capable of competently performing a wide range of distinct tasks, and can be integrated into a variety of downstream systems. The model itself is a component: it becomes an AI system once combined with a user interface. GPAI models are governed by Articles 51–56; systems built on top of them by Article 50.
- Code of PracticeA code of practice — within the AI Act framework, best known in the form of the GPAI Code of Practice — is a legally non-binding soft-law document that serves as a voluntary means of demonstrating compliance pending the emergence of a harmonised standard. Signing it makes it easier for a provider to prove compliance, but non-signature is not, by itself, an infringement, provided the provider can present alternative, equivalent proof.
Related questions in the knowledge base (Hungarian)
- Mit jelent a rendszerszintű kockázatú AI-modell, és mi a 10^25 FLOP küszöb?Az 51. cikk (2) bekezdése szerint rendszerszintű kockázatot kell vélelmezni, ha a modell tanítására fordított halmozott számításigény meghaladja a 10^25 FLOP-ot. A szolgáltatónak két héten belül értesítenie kell a Bizottságot, és teljesítenie az 55. cikk négy többletkötelezettségét. A vélelem megalapozott érveléssel megdönthető, a Bizottság viszont elutasíthatja.
- Mit jelent a GPAI Code of Practice aláírása, és kötelező-e csatlakozni?A gyakorlati kódex jogilag nem kötelező soft law. Aláírása a megfelelés önkéntes bizonyítási eszköze az 53. cikk (4) és az 55. cikk (2) alapján, a harmonizált szabvány megjelenéséig. A nem-aláírás önmagában nem jogsértés, de a szolgáltatónak alternatív, egyenértékű bizonyítást kell felmutatnia, egyedi hatósági értékelés mellett.