Detecting undisclosed LLM-generated content in parliamentary texts

In three linesarXiv study detecting undisclosed LLM-generated content in UK and Swedish parliamentary texts. An interpretable classifier trained on pre-LLM texts and their generated versions shows steady increase in undisclosed LLM use since 2022 in both parliaments.

Summary generated by Claude — human-verified