Data Pseudonymization in the Generative Artificial Intelligence Environment

This paper deals with data pseudonymization when using generative artificial intelligence, specifically ChatGPT. The aim of the research was to experimentally verify whether ChatGPT can maintain the protection of pseudonymized data by not maintaining the link between the anonymized part and personal data. Four requests were tested in the experiment, in which ChatGPT was gradually provided with various combinations of pseudonymized and complete data about customers and their orders. The results showed that ChatGPT does not store previous inputs and is not able to reconstruct personal data retroactively if they were not explicitly provided to it in each request. Based on the results obtained, it can be assumed that the use of ChatGPT when processing pseudonymized data does not pose a risk of violating the protection of personal data. This research seeks to contribute to the discussion on the security of generative artificial intelligence in processing sensitive data and confirms the effectiveness of pseudonymization as a mechanism for protecting personal data.