AI disclosure rules are solving the wrong problem, according to a new paper posted to arXiv's AI section.
The paper argues that questions like whether AI was used, whether that use was disclosed, and whether hidden use can be detected miss a deeper issue: judgment itself can become unowned, with no human or institution left accountable for it. The authors examine two cases. In AI-assisted peer review, contribution dissolution can spread responsibility for a critique across reviewer and tool until no one is clearly on the hook for it. In creative work, fear of losing credit for a piece can discourage people who used AI from disclosing it honestly. Both cases, the paper argues, expose the limits of disclosure rules and provenance records as fixes.
That reframing matters because most current AI policy debates, including disclosure checkboxes and detection tools, focus on managing AI use itself rather than on who stands behind the resulting decision. The paper's proposed fixes, distinguishing the role AI actually played, flagging which judgments need a clear human owner, and easing the default penalty for disclosing AI involvement, aim at accountability rather than surveillance.
Writing a disclosure policy is easy. Deciding who owns the judgment behind an AI-assisted decision is the harder problem nobody has solved yet.