The technology is already in use. The question is whether institutions can catch up.
Quietly and without formal decree, artificial intelligence has crossed a threshold in the world of scientific knowledge-making: more than half of all peer reviewers now enlist these tools in evaluating the research that shapes our collective understanding. A report from publisher Frontiers, drawn from nearly 1,700 researchers across the globe, reveals that adoption has outpaced governance — that practice has, as it so often does, run ahead of principle. The moment calls not for alarm but for deliberate reckoning, as the institutions entrusted with scientific integrity must now decide whether they will shape this transformation or simply follow in its wake.
- AI has already crossed the majority threshold in peer review, with 53% of reviewers using it globally and 87% of early-career researchers leading the charge — the shift is not coming, it has arrived.
- The absence of consistent policy across journals and borders creates a fractured landscape where the same AI-assisted review might be accepted in one venue and flagged as misconduct in another.
- Researchers are not resisting this change — they are asking for guidance, signaling that the hunger for clear rules is as strong as the appetite for the tools themselves.
- Frontiers has proposed a concrete framework: mandatory disclosure of AI use, universal access to vetted tools, training programs, and strengthened oversight to prevent the integrity gap from widening.
- The deeper stakes lie ahead — AI capable of detecting methodological flaws and verifying reproducibility could either fortify the scientific record or, without governance, quietly erode it.
More than half of the researchers who evaluate scientific papers for publication are now using artificial intelligence to assist them — a finding drawn from Frontiers' interviews with 1,645 active researchers worldwide. The shift has happened with remarkable speed and, largely, without formal rules to govern it.
For now, the uses are practical: drafting written evaluations, summarizing findings, polishing language. But Elena Vicario, who leads research integrity at Frontiers, points toward a more consequential horizon — AI tools capable of identifying methodological flaws, verifying reproducibility, and exposing deeper problems in study design. The capability exists. The governance does not yet match it.
The adoption numbers reveal an uneven but sweeping momentum. Early-career researchers are the most enthusiastic adopters at 87%, followed by researchers in China at 77% and Africa at 66%. These are not fringe figures — they represent a fundamental change in how science gets evaluated, unfolding largely outside formal institutional guidance.
What researchers say they need most is clarity. The whitepaper found genuine enthusiasm for expanded AI use alongside a strong desire for consistent policy — something that would define acceptable practice across journals and borders. Currently, rules vary wildly or don't exist at all, creating confusion and potential inequity for those with access to more powerful tools.
Frontiers' proposed framework calls for disclosure requirements, researcher training, stronger oversight, and equitable access to trustworthy AI tools beyond wealthy institutions. CEO Kamila Markram frames the moment as an opening — for better quality, broader collaboration, and greater inclusion of historically underrepresented research communities — but insists that realizing it demands coordinated action across publishers, universities, funders, and developers.
The whitepaper is ultimately a call for institutions to catch up with their own researchers. The technology is already woven into scientific practice. The question is whether the structures that safeguard knowledge can move quickly enough to ensure AI becomes a force for integrity rather than a quiet threat to it.
More than half of the people who review scientific papers for publication are now using artificial intelligence to help them do their work. That's the finding from a new report by Frontiers, a research publisher, based on interviews with 1,645 active researchers around the world. The shift has happened quietly and quickly—so quickly that the publishing industry is scrambling to figure out what rules should govern it.
Right now, reviewers are using AI mostly for straightforward tasks: drafting their written evaluations of papers, summarizing research findings, cleaning up their language. But the real opportunity, according to Elena Vicario, who directs research integrity at Frontiers, lies in what comes next. AI could help catch methodological flaws, verify reproducibility, and surface deeper problems in how studies are designed. The tools exist. The question is how to use them responsibly, and who gets to decide.
The adoption numbers tell a story of uneven momentum. Among early-career researchers—those still building their reputations and establishing their place in science—87% are already using AI in peer review. In China, 77% of researchers surveyed said they use these tools. In Africa, the figure is 66%. These are not marginal numbers. They represent a wholesale shift in how scientific work gets evaluated, happening without much formal guidance from above.
What researchers say they want is clarity. The whitepaper found broad enthusiasm for using AI more extensively, but also a clear hunger for consistent policy recommendations that would tell them what's acceptable and what crosses a line. Right now, different journals have different rules, or no rules at all. A reviewer in one country might be using AI in ways that would be flagged as problematic elsewhere. The lack of alignment creates confusion and, potentially, unfair advantages for those with access to better tools.
Frontiers has proposed a framework to address this gap. The recommendations include requiring reviewers to disclose when and how they've used AI, building training programs so researchers understand both the capabilities and the limitations of these tools, strengthening oversight mechanisms to catch misuse, and making sure that trustworthy AI tools are available to researchers everywhere, not just those at wealthy institutions. The company also emphasizes the importance of data provenance—being able to trace where information came from and how it was processed.
Kamila Markram, Frontiers' CEO, frames this moment as an opportunity rather than a crisis. AI is transforming how science gets written and reviewed, she said, opening possibilities for better quality, wider collaboration, and greater participation from researchers in parts of the world that have historically been underrepresented in global science. But realizing that potential requires the whole ecosystem—publishers, universities, funding agencies, the companies building the tools—to move together toward aligned policies and responsible governance.
The whitepaper is, in effect, a call for the research world to catch up with itself. The technology is already in use. The question now is whether the institutions that oversee science can establish rules and training systems fast enough to ensure that AI strengthens the integrity of the scientific record rather than undermining it.
Citas Notables
AI is already improving efficiency and clarity in peer review, but its greatest value lies ahead. With the right governance, transparency, and training, AI can become a powerful partner in strengthening research quality.— Elena Vicario, Director of Research Integrity at Frontiers
AI is transforming how science is written and reviewed, opening new possibilities for quality, collaboration, and global participation. This whitepaper is a call to action for the whole research ecosystem to embrace that potential.— Kamila Markram, CEO and Co-founder of Frontiers
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that peer reviewers are using AI? Isn't peer review just peer review?
Because peer review is how science polices itself. If half the people doing that work are using tools that no one has agreed on how to use, you've got a system that's changing without anyone really steering it.
But the numbers sound positive—87% of early-career researchers using AI, high adoption in China and Africa. Isn't that democratizing science?
It could be. But only if those researchers have access to the same quality tools and the same training. Right now, a researcher at a well-funded university might be using AI in ways a researcher in a less wealthy institution can't even access. That's not democratization; that's creating new hierarchies.
What's the actual risk here? What goes wrong if we don't establish governance?
You could have reviewers using AI to evaluate papers without disclosing it. You could have tools making mistakes that no human catches because everyone assumes the AI is reliable. You could have bias baked into the tools affecting which research gets approved and which doesn't. And you could have the whole system losing credibility if people find out the scientific record was shaped by undisclosed, ungoverned AI use.
The report says AI could improve reproducibility and catch methodological flaws. That sounds like a genuine benefit.
It is. That's why this isn't about stopping AI use. It's about doing it transparently, with training, with oversight. The tools are good. The problem is we're using them in the dark.
What would good governance actually look like?
Transparency first—reviewers have to say when they used AI and how. Training so people understand what these tools can and can't do. Oversight to catch misuse. And making sure the tools are available to everyone, not just researchers at rich institutions. Without that last part, you're just automating existing inequalities.