EU AI Act: first systemic-risk evaluation deadline passes
Providers of the most powerful AI models had to submit their first formal risk reports to the EU AI Office by September 15.
Illustration · AI-generated (AI IN LIFE)
At a glance
- Active enforcement of the EU AI Act since August 2, 2026
- Deadline for first systemic-risk evaluations: September 15, 2026
- Threshold: general-purpose models trained above 10^25 floating-point operations
- Reviewed: red-teaming methods, energy consumption, copyright training summaries
- Roughly 40 new positions posted at the EU AI Office to strengthen enforcement
The European Union has reached another key milestone in enforcing its AI Act: since August 2, 2026, the EU AI Office and national authorities have been actively enforcing the rules. On September 15, a first concrete deadline passed, by which providers of the most powerful AI models had to submit their first formal systemic-risk evaluations.
The rule applies to providers of general-purpose AI models whose training exceeds a defined computing-power threshold of 10^25 floating-point operations, a bar currently cleared only by the industry's largest models. The EU AI Office is reviewing the red-teaming methodologies used, model energy-consumption disclosures, and compliance with the standardized copyright training-data summary template published in July.
To handle the growing volume of reviews, the office is significantly expanding its staff: roughly 40 new positions across technical, legal, and operational roles were posted, with expressions of interest due by September 8.
FAQ
What counts as a systemic risk under the AI Act?
It refers to potential large-scale harms that especially capable general-purpose AI models could cause, such as misuse, loss of control, or wide-reaching societal impact.
Which companies are affected?
Providers of general-purpose AI models whose training exceeds the 10^25 operations threshold — in practice, only the industry's largest models.
What happens in case of violations?
The EU AI Office and national authorities conduct technical audits; the AI Act provides for substantial fines in case of violations.
