Beyond AI disclosure: Claim accountability and responsible research in scholarly publishing

Abstract

Generative artificial intelligence is now part of everyday scholarly practice. It can help researchers polish text, debug code, translate text, and support deductive coding under human supervision. But it can also produce fluent, competent-looking content that no one is positioned to defend. Journal and publisher policies have largely responded to AI through prohibition, mandatory disclosure, or validity-only permissiveness. Each regulates AI use in the production of manuscripts rather than the defensibility of scholarly claims, leaving the central accountability problem intact. We argue that the right unit of governance is the claim, not the tool. Responsible research with AI requires a named human who can reconstruct and defend each claim that enters the scholarly record. We develop a two-threshold framework, in which the confidentiality threshold protects entrusted material in peer review and editorial deliberation, while the judgment threshold protects human responsibility for the reasoning behind scholarly claims. We then translate the framework into role-based self-assessment for authors, reviewers, editors, and publishers, and identify the institutional anchors that prevent it from decaying into boilerplate. AI is the stress test of scholarly publishing. It has exposed weaknesses in how we evaluate, produce, and protect knowledge, and no gain in efficiency justifies a published claim that no human can defend.