A recent large-scale audit of dataset licensing and attribution in artificial intelligence (AI) has revealed significant issues in the field. Published in Nature Machine Intelligence, the audit examined over 1,800 AI training sets and found a 'crisis of misattribution,' with more than 70% of datasets having license omissions and over 50% containing license errors. This research highlights the crucial role of data attribution in AI development, emphasizing its importance for fairness, accountability, and improving model performance. The findings underscore the need for better data provenance as sources increasingly restrict open access to information.