The European Union's Artificial Intelligence Office has taken its most decisive enforcement step to date, publishing the standardized compliance templates under the EU AI Act for General Purpose AI models.
The framework establishes tiered obligations: all baseline LLMs must publish comprehensive training data summaries, while frontier systems exceeding 10^25 floating-point operations must undergo third-party adversarial red-teaming.
Tech industry representatives from Washington and Beijing confirmed active working groups to align safety evaluations with Brussels standards.