Determine whether a GPAI model has systemic risk under AI Act Article 51 and notify the AI Office under Article 52

Documented steps

1. Read Article 51: a GPAI model is presumed to have high-impact capabilities, and therefore systemic risk, when the cumulative compute used for its training measured in floating point operations is greater than 10^25. Text: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-51
2. Check the second route to classification in Article 51(1)(b): the Commission may designate a model as systemic-risk on its own initiative or following a qualified alert from the scientific panel, applying the Annex XIII criteria (capability benchmarks, number of business and end users, market reach, tool and data access, scalability). Staying below 10^25 FLOP does not put you out of scope.
3. Project the training-run compute before and during training, not only after it completes — the Article 52(1) clock can start once it becomes known that the threshold will be met.
4. Notify the Commission without delay and in any event within two weeks of the requirement being met, or of it becoming known that it will be met. Text: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-52
5. Submit the notification through the EU SEND platform, which the Commission's GPAI guidelines identify as the channel for GPAI submissions to the AI Office including Article 51(2)/52(2) notifications. Access details: https://digital-strategy.ec.europa.eu/en/policies/guidelines-gpai-providers
6. If you consider that the model exceptionally does not present systemic risk despite crossing the threshold, submit sufficiently substantiated arguments with the notification. Notify first — the rebuttal does not suspend the two-week deadline.
7. If the Commission rejects the rebuttal or designates the model, implement Article 55 obligations in addition to Article 53: state-of-the-art model evaluation including adversarial testing, systemic-risk assessment and mitigation, tracking and reporting of serious incidents, and adequate cybersecurity protection for the model and its physical infrastructure.
8. Monitor the Commission's published list of GPAI models with systemic risk to confirm your model's status and to see peer designations.
9. If designated, note that a reassessment request can only be made after six months from designation, and must present objective, detailed and new reasons.
10. Engage the AI Office early where a run is expected to approach the threshold rather than relying solely on the post-hoc two-week notification.