Researchers propose Humanly, a platform to document and audit human-AI collaborative writing sessions

Researchers from multiple institutions introduced Humanly, a writing platform designed to make the writing process itself the primary evidence for evaluating authorship and collaboration. The platform allows users to configure writing environments for personal documents or assigned tasks, recording both keystrokes and in-platform AI assistance during drafting.

Humanly packages completed writing sessions into sealed 'writing certificates' that include configuration-aware anomaly behavior review, enabling verification of the writing process for scenarios such as course assignments, peer review, and personal certification. The platform is intended to address limitations in existing process-tracking tools, which often rely on coarse activity records tied to host-document histories.

A user study conducted by the researchers found that Humanly was helpful across roles, including writers, reviewers, and evaluators. Additionally, a red-teaming study demonstrated that the platform's Humanly Typing Detector could distinguish human hand typing from automated typing, providing a technical mechanism to audit authorship claims.

The authors include Shenzhe Zhu, Haoqian Zhang, Xu Yang, Jingyu Tang, Yi Nian, Xiaoxue Du, Shu Yang, Alex Pentland, Joachim Baumann, and Jiaxin Pei, and the work is presented as an arXiv preprint (arXiv:2607.21758) submitted on July 23, 2026.