AI and the research system: report round-up
Three new publications in the past week have explored the question of AI’s impact on the research system from three different angles.
Oxford University Press has surveyed researchers about AI use, with a focus on disclosure when submitting articles to journals; Clarivate has released its annual polling on library AI adoption; and the Innovation and Research Caucus has recommendations around how UKRI might incentivise the development of AI skills. Below, we pick out the main highlights.
AI disclosure
Oxford University Press has updated its journal guidelines for both editors and authors concerning the “responsible, appropriate, and transparent use of AI.” The move comes alongside the release of its Disclosing AI use in academic research report, based on a survey of around 2,600 researchers around the world, which found that just shy of three-quarters of respondents felt there was a “general lack of clarity” about how they should disclose the fact that they have used AI tools when it came to submitting work to a journal.
One finding contrasts frequency of AI use by discipline, suggesting that such tools are seeing less pick-up in the arts and humanities:
The survey results also indicated that early-career researchers were somewhat more likely to use AI tools in their work than those in mid- or late-stage. The most common forms of AI employed were chatbots, translation tools, and plagiarism checkers – three-quarters of those polled said they had not used any agentic AI in the last year.
When it came to manuscript submission, four-in-ten of those who had recently published a journal article said that the platform had not asked them to declare AI use:
For those who had published a book in the last year, this rose to 52 per cent reporting that they were not asked to declare AI use (from a smaller sample of 789 respondents). For grant applications, the figure was 61 per cent (535 respondents).
OUP has said that its new guidelines will emphasise transparent disclosure and human accountability. It also references ongoing work supported by the International Science Council and others to develop a global reporting standard, which is currently surveying the research community for input ahead of final publication.
Academic libraries
Clarivate’s Pulse of the Library survey for 2026 assesses the progress of academic and other libraries in implementing AI, finding “cautious AI adoption, but with no improvement in confidence over 2025 and few gains in implementation.” The survey is based on 1,876 responses from librarians around the world, two-thirds of whom were based in academic library contexts.
The picture of implementation is indeed largely unchanged on previous iterations of the survey, though the proportion of academic librarians who said their library was actively or moderately deploying AI has risen over the last two years:
Libraries in the United States are found to be somewhat more cautious about AI use than elsewhere in the world, in particular mainland China and other locations in Asia.
Where academic librarians were asked about their primary concerns with adopting or scaling AI technology deployment, the most frequently seen responses related to academic integrity, privacy, and budget constraints, though various others were well-represented:
UKRI and AI skills
Finally, a new UKRI-commissioned study from the Innovation and Research Caucus looks at what UKRI should be doing to promote AI skills among researchers. It makes the case that the funder has the opportunity to “bring coherence, visibility and rigour to a fragmented AI skills landscape.”
The report, based on desk research alongside a small number of interviews and workshops, first outlines the barriers to developing, attracting and retaining AI skills within the research system. These are found to include: the difficulty in staying abreast of rapid development; a “crowded and fragmented” training market making it hard to identify credible training provision; retention challenges; concerns about bias and inclusivity in AI-enabled research and innovation, in particular related to the under-representation of women among AI developers; and lack of access to data and compute for training purposes.
The authors set out recommendations for how UKRI can better build AI skills within the research and innovation workforce. First and foremost is for an AI Skills Champion role to be resourced, ideally one per research council to ensure “close alignment with disciplinary needs.”
It is also suggested that UKRI further enhance its responsible research and innovation guidance, and “use funding levers” to support adherence to responsible AI use – this might see grant applicants required in some way to demonstrate compliance, or institutions expected to provide relevant training.
Research Agenda is interested in hearing from readers about how AI is impacting on research in specific disciplines. Read the first in our series here, and drop us a line to share your thoughts.