The development of AI models in joint projects yields improved results because multiple companies contribute their data. The situation becomes critical, however, if a partner withdraws and requests the return of its data. With federated unlearning, Fraunhofer researchers have developed a method for precisely removing data from decentralized AI models.