Detecting undisclosed LLM-generated content in parliamentary texts

This paper presents a study on the detection of undisclosed LLM-generated content in parliamentary texts from the UK and Sweden, highlighting the importance of transparency in AI use within governmental writing. An interpretable text classifier was developed, trained on both pre-LLM parliamentary texts and their LLM-generated counterparts, revealing a notable increase in undisclosed AI usage in parliamentary documents since 2022. This work underscores the need for clear guidelines on AI disclosure in legislative contexts, which is critical for maintaining public trust and accountability.