The metaphor of the open seminar is misleading. In an open seminar, the presenter stake a claim to the work, even if incomplete. In this case, the work was in private repositories. A more precise methodology would be the AI overhearing the authors discussing in their office and using that as base for a proof. This is even sneakier because there has been no public claim of the method or results by the authors.
Edited after a commenter pointed out that no sleep + autocorrect = random words