AI Safety Lead Questions Dataset Download Origins

Maksym Andriushchenko, principal investigator of the AI Safety and Alignment Group at ELLIS Institute Tübingen and MPI-IS, posted about PostTrainBench trajectories. He noted they were downloaded more than 35k times last month and wondered whether the activity came from real people or rogue AI agents. The dataset is available from aisa-group on Hugging Face, a site dedicated to open source AI development. The post included a screenshot of the dataset page.