Two academic papers released over the past week claim rapid progress toward fully autonomous scientific research. A survey led by Shanghai AI Lab, titled "From AI for Science to Agentic Science," maps the field’s trajectory from task-specific tools to AI agents capable of independently designing and executing studies. The authors catalogue outstanding hurdles—reproducibility, novelty validation and transparency—and propose benchmarks such as a cross-disciplinary Nobel-Turing Test to gauge machine-generated discoveries. Complementing the survey, an interdisciplinary team spanning the University of Queensland and other institutions posted "Virtuous Machines: Towards Artificial General Science" on arXiv on 19 August. The paper describes a domain-agnostic system that autonomously generated hypotheses, recruited 288 online participants, analysed data and drafted full manuscripts for three cognitive-psychology experiments in roughly 17 hours each, at a marginal cost of about US$114 per study. The authors say the manuscripts met methodological standards comparable to experienced human researchers, although the AI showed limitations in conceptual nuance. Together, the publications argue that agentic AI could accelerate discovery by handling the entire research pipeline, but they emphasise the need for new governance frameworks to assign credit, ensure ethical oversight and verify results before such systems are deployed at scale.