The VeriDark benchmark contains several large-scale authorship verification and identification datasets, which should facilitate research into authorship analysis generally and enable building tools for the cybersecurity domain in particular.
The datasets are detailed here, while the BERT-based baseline code can be found here.