The AI Act of the EU is intended to regulate the use of artificial intelligence in the member states. Like all EU regulations, it is binding in 24 official languages; with a text of around 80,000 words (based on the German version), ensuring stringent consistency across all languages is difficult. Particularly taking into account the technical context of the criteria ‘accuracy’ and ‘robustness’, the present corpus study analyses the differences in the official language versions, and the question of whether these differences have an influence on technical requirements and legal assessments.