A recent paper posted on arXiv asks how copyright governance should work in the era of generative AI. The authors note that large language models are trained on collections that may include scholarly works, yet authors typically have no work-level information about whether or how their own publications were used.

This information gap sits at the center of the paper's inquiry. Without knowing which works entered a training set, authors cannot meaningfully exercise rights or seek remedies. The paper frames copyright governance as a question of responsibility: what should institutions, model builders, and policymakers do when usage is effectively invisible to the people who created the works?

The abstract does not lay out specific proposals, but it signals a focus on governance structures rather than purely technical fixes. The significance lies in shifting attention from whether training on copyrighted works is legal to how the process can be made more transparent and accountable to authors.