Abstract
What are the possibilities of auditing artificial intelligence, and how can they be realized in theory and in practice? The global use of artificial intelligence is highly increasing. The digital forensic community therefore needs forensic methods to audit internal decisions to ensure their reliability, especially in safe-critical environments. Furthermore, forensic readiness measures are equally necessary in the event of an attack on AI systems. The continual learning framework Avalanche is used to construct an application-oriented scenario as the foundation to store weight values and weight changes in a log file while training. The development of the changes as well as the resulting memory size with varying hidden layer and neuron numbers provides conclusions about the applicability of the method, which needs further research in the future.