European UnionAI Act, Articles 51 and 52 (classification of general-purpose AI models with systemic risk and the notification procedure)
Regulation (EU) 2024/1689, Articles 51 and 52, with Annex XIII
In force since .
An AI governance rule binding public and private bodies.
- How the hook was established
- express
- What makes it apply
- Market targeting
- Obligation class
- Reporting
- Audit expectation
- none
As of .
What it requires
- If you are the provider of a general-purpose AI model that has high impact capabilities, which Article 51(2) presumes when the cumulative amount of computation used for its training is greater than 10^25 floating point operations, notify the Commission without delay and in any event within two weeks after that requirement is met or it becomes known that it will be met, with the information necessary to demonstrate that the requirement has been met.
- With your notification you may present sufficiently substantiated arguments that, exceptionally, although your model meets that requirement, it does not present systemic risks because of its specific characteristics and so should not be classified as a model with systemic risk. If the Commission concludes that the arguments are not sufficiently substantiated and you were not able to demonstrate this, it rejects them and your model is considered a model with systemic risk.
- If the Commission designates your model as presenting systemic risks, ex officio or following a qualified alert from the scientific panel, on the basis of the Annex XIII criteria, you may make a reasoned request for reassessment that contains objective, detailed and new reasons that have arisen since the designation decision, at the earliest six months after that decision and, where the Commission maintains the designation, at the earliest six months after the decision that maintains it.
- If your model was placed on the market before , take the necessary steps to comply with the obligations laid down in the Regulation by .
These lines are LexLint's reading of the law's text, as of .
If you get it wrong
Private right of actionNo
Penalty structure
Article 101(1): the Commission may fine a provider of a general-purpose AI model up to 3% of its annual total worldwide turnover in the preceding financial year or EUR 15,000,000, whichever is higher, where it finds that the provider intentionally or negligently infringed the relevant provisions of the Regulation (point (a)), failed to comply with a request for a document or for information under Article 91 or supplied incorrect, incomplete or misleading information (point (b)), failed to comply with a measure requested under Article 93 (point (c)), or failed to make access to the model available for an evaluation under Article 92 (point (d)).
- Rule
- Higher of
- As of
- Currency
- EUR
- Fixed cap
- 15,000,000
- Turnover percentage cap
- 3
Who enforces it
Enforcement body
The European Commission, which holds exclusive powers to supervise and enforce Chapter V of the Regulation and entrusts the implementation of those tasks to the AI Office.
Settledness
- As of
- Open questions
- When does it 'become known' under Article 52(1) that a general-purpose AI model will meet the high impact capabilities condition of Article 51(1), point (a), for example while training is still under way and the cumulative computation of Article 51(2) has not yet been reached?
What it makes you log
Logging duty
The notification must include the information necessary to demonstrate that the requirement has been met, so a provider needs a record from which to show it. Article 52 names no log, record or retention period of its own, and the technical documentation a provider keeps for the model is dealt with in Article 53(1), point (a).
- Kind
- Implicit
- As of
- Provision
- Article 52(1)
What this law does
A general-purpose AI model is classified as a model with systemic risk if it has high impact capabilities, evaluated on the basis of appropriate technical tools and methodologies, including indicators and benchmarks.
A model is also classified as a model with systemic risk if, based on a decision of the Commission taken ex officio or following a qualified alert from the scientific panel, it has capabilities or an impact equivalent to those high impact capabilities, having regard to the criteria set out in Annex XIII. A model is presumed to have high impact capabilities when the cumulative amount of computation used for its training, measured in floating point operations, is greater than 10^25.
The Annex XIII criteria for equivalent capabilities or impact are the number of parameters of the model, the quality or size of the data set, the amount of computation used for training, the input and output modalities, the benchmarks and evaluations of capabilities, the impact on the internal market (presumed when the model has been made available to at least 10,000 registered business users established in the Union) and the number of registered end-users.
The provider of a model that meets the high impact capabilities condition must notify the Commission without delay and in any event within two weeks after that requirement is met or it becomes known that it will be met. The notification that Article 52(1) requires must include the information necessary to demonstrate that the requirement has been met.
If the Commission becomes aware of a model presenting systemic risks of which it has not been notified, it may decide to designate it as a model with systemic risk.
The provider of a model that meets the high impact capabilities condition may present, with its notification, sufficiently substantiated arguments that, exceptionally, although the model meets that requirement, it does not present systemic risks because of its specific characteristics and so should not be classified as a model with systemic risk.
Where the Commission concludes that arguments submitted under Article 52(2) are not sufficiently substantiated and the provider was not able to demonstrate that the model does not present systemic risks, it rejects the arguments and the model is considered a model with systemic risk. The Commission may designate a general-purpose AI model as presenting systemic risks, ex officio or following a qualified alert from the scientific panel, on the basis of the criteria set out in Annex XIII.
Upon a reasoned request of a provider whose model has been designated, the Commission takes the request into account and may decide to reassess whether the model can still be considered to present systemic risks on the basis of the Annex XIII criteria.
A request for reassessment must contain objective, detailed and new reasons that have arisen since the designation decision, and a provider may make it at the earliest six months after the designation decision or, where the Commission maintains the designation after reassessment, at the earliest six months after that decision.
The Commission publishes and keeps up to date a list of general-purpose AI models with systemic risk, without prejudice to the need to observe and protect intellectual property rights and confidential business information or trade secrets. Article 113 applies Chapter V, which contains Articles 51 and 52, from .
A provider of a general-purpose AI model placed on the market before has until to take the necessary steps to comply with the obligations laid down in the Regulation. The Commission may fine a provider of a general-purpose AI model up to 3% of its annual total worldwide turnover in the preceding financial year or EUR 15,000,000, whichever is higher, where it finds that the provider intentionally or negligently infringed the relevant provisions of the Regulation.
Article 101 applies from , the date from which the Regulation applies, because Article 113 excepts it from the earlier date given to Chapter V.
When LexLint raises it
When your app profile says your app trains models.