AI for Science: Provenance, Peer Review, and Scientific Independence

AI is entering scientific workflows through attributable access to approximately 20 million paywalled papers, automated literature work, and peer-review assistance. Early evidence shows potential utility but also severe over-acceptance, weak calibration, generative misrepresentation, and risks that shared tools homogenize methods and research questions. Licensing, retrieval quality, human oversight, and safeguards for scientific diversity remain unsettled.