Breaking the shackles of confidentiality: Why AI should “free” scientific knowledge

The scientific community is currently debating the ethical implications of using generative artificial intelligence (AI) in peer review at a time when manuscript volumes, reviewer burden, and the pace of AI-assisted knowledge production are increasing. Traditional governance models, emphasizing strict manuscript confidentiality and the protection of intellectual property, largely prohibit the use of AI tools by reviewers. This topic piece argues that while confidentiality is a foundational element of academic trust, treating all unpublished manuscripts as absolute secrets can inadvertently delay the dissemination of socially valuable knowledge. The prevailing binary, either uncompromising secrecy or unregulated AI-mediated openness, is an inadequate ethical framework for modern scholarly communication. Rather than relying on ad hoc prohibitions, we propose a shift toward deliberate institutional design. The ethical integration of AI into peer review requires the development of secure, auditable AI infrastructures that do not use submitted manuscripts or review reports for external model training. By adopting author-consented, risk-sensitive models of AI-assisted review, the scientific community can preserve reviewer accountability while exploring controlled forms of openness and accelerating knowledge evaluation for societal benefit.