EU AI Act Art. 51 — Classification of general-purpose AI models with systemic risk
Article 51 decides which general-purpose AI models count as presenting systemic risk: those with high-impact capabilities, or trained above a compute threshold of 10²⁵ FLOPs — triggering the strictest GPAI obligations.
At a glance
What this article requires
- A GPAI model has systemic risk if it has high-impact capabilities, assessed via benchmarks and tools.
- It is also presumed systemic when cumulative training compute exceeds 10²⁵ FLOPs.
- The Commission can designate models as systemic risk ex officio based on Annex XIII criteria.
- Systemic-risk classification triggers the strictest GPAI obligations (Art. 55 and related rules).
- The GPAI chapter applied from 2 August 2025.
Scope
Who this applies to
Providers of general-purpose AI models — mostly foundation-model labs, but also companies fine-tuning or building on open-weight models that cross the thresholds.
Obligations
What you must actually do
Self-assess your compute
Track cumulative training compute honestly; crossing 10²⁵ FLOPs makes you systemic unless you can argue exceptional characteristics.
Notify the Commission
Providers of models that meet the threshold must notify the Commission within two weeks of meeting — or expecting to meet — it (Art. 52).
Track the designation list
The Commission publishes the list of systemic-risk models — monitor it, and check which open models you build on appear on it.
Action plan
Practical first steps
- 1
Add compute and capability tracking to your training runs.
- 2
Prepare the technical capability documentation before you hit the threshold, not after.
- 3
Watch the Commission's GPAI guidance and the systemic-risk list for your models and the ones you depend on.
Penalty exposure
Violations sit in the GPAI penalty regime: up to €15 million or 3% of global annual turnover, with special provisions under Art. 101.
FAQ
Questions about Art. 51
What does 10²⁵ FLOPs mean in practice?
It is a very high bar — roughly frontier-scale pretraining. Most models that most companies build or fine-tune are nowhere near it; the threshold targets the largest foundation models.
Can I be systemic without crossing the FLOPs threshold?
Yes — the Commission may designate a model as systemic risk based on high-impact capabilities even below the compute threshold.
Sources
Citations & further reading
Related
More article explainers
Wondering which articles apply to your AI?
Describe your system in the free Risk Scanner and get a preliminary risk read with the obligations that likely apply — in seconds.
Check my use casePreliminary EU AI Act clarity summary. Not legal advice.