Blog post by Benjamin White
Originally published in a shortened version on 11 March 2026 by Research Professional News
Universities make a uniquely important contribution to the UK’s research intensive economy, both through generating new ideas and insights through basic research, and then through their application, for example in the form of research–industry collaborations, spinouts and more.
In order to maximise productivity, like almost every other sector, the research sector is looking at how to make the most of AI technologies. Closely connected to this, UK researchers are also working to develop new models and applications of AI that can support numerous sectors of the economy including the NHS, the environment, agriculture etc.
In this context, as we see with the ongoing furore around Palantir, it is a cause for concern that we are so dependent on US technology in an increasingly hostile geo-political environment. Just as crucially important datasets were quietly altered in the early days of the Trump Administration, what is to say that US platforms and model-providers will not also find themselves pressured to stymy the information and services they provide, or only to provide them under unacceptable conditions.
Faced with this, the UK needs to focus investment and regulation on developing an independent digital infrastructure for AI, namely work towards “AI digital sovereignty”. This would safeguard academic autonomy while supporting the upstream innovation on which a research-intensive economy depends.
Achieving AI digital sovereignty will involve multiple pillars of action. Turning first to skills, an obvious need is around programming. We can make make the most of these skills if research organisations also open up opportunities for interdisciplinary engagement. Beyond this, we also need a high level of information literacy among researchers and students, through building critical understanding of the strengths and weaknesses of particular AI models. This will help ensure effective, informed, accurate, transparent and ethical use of AI.
Second is a need to invest in technical infrastructures. Intellectual sovereignty and freedom of expression can be compromised by institutional dependence on commercial LLMs. Documented biases and distortions such as those from the US (e.g. Elon Musk owned GROK ranting about “white genocide” in his own country South Africa) and China (e.g. DeepSeek’s twisting of information on Taiwan and Tiananmen Square) illustrate this need. We need to encourage open and non-biased models as well as home-grown ones that can support research.
Connected to this are questions about how far research-intensive universities can be digitally autonomous in AI when they rely on computing capacity based elsewhere. Currently, 70% of global cloud infrastructure is owned by Google, Amazon and Microsoft, and the fastest and most advanced public cloud centres are in the US and China. Until governments around the globe manage to provide sustainable alternatives, the extent to which universities can rely on these centres will depend not only on their ability to pay, but also their willingness to place their institution’s activities under the extraterritorial powers of the US and Chinese law enforcement agencies.
Third, there is the development of policy, notably at the level of institutions and disciplines around what uses of AI are permissible, and subsequently what machine learning software complies with this. Clarity here will offer a valuable time-saver, helping to make the UK a more attractive place to do research.
No AI Digital Sovereignty without Data
A crucial strand of AI digital sovereignty however is the ability of our institutions to freely use the oil that fuels the AI machines – data.
There are two key, interconnected aspects to this – the ability of the UK research sector to draw on third-party data, and how far the data it produces supports responsible AI development and application.
Looking at the first, as with technical infrastructures, geography and regulation have a very clear influence on the extent to which universities may use third party data for model training and subsequent uses. In the context of AI, depending on the area of research, this often means heterogenous datasets such as webpages, books, newspapers, journal articles, clinical trial data and more. Crucially, when we are talking about use by a university, the NHS or any other research performing organisation, this includes use within the context of research-industry collaborations and spin-outs which, as established at the beginning of this article, are an integral part of the UK’s research ecology.
Whereas the copyright system in countries the US, Japan, South Korea and Singapore provide that once a university has legitimate access to such works for human reading there are no legal barriers to reusing them for AI, including in collaborative projects with private or public partners, this is not the case in the UK. Here, in order to share data between institutions, or NHS trusts or private companies, UK based entities must seek a new licence despite having legal access to the data already – something that any licensing professional will tell you is not possible at any scale. And of course, the big data that fuels AI demands scale.
Despite the clear benefits for the UK research sector – and so the wider economy – that would accrue from allowing research-oriented organisations to share and analyse legitimately accessed data more freely as many of our competitors can, in the context of the ongoing debates around copyright and AI the government seems minded to leave copyright laws where they are. Moreover, the government seems blind to the fact AI impacts anyone else but Big Tech and the creative industries. For a government interested in growth, having just released an AI and Science strategy this is confusing to say the least.
The situation is particularly absurd when it comes to the question of whether universities and researchers can leverage their own academic outputs for AI. Many academics routinely assign their copyright to publishers, which as pointed out above means in the UK another licence will be required to use (often publicly funded research) for AI purposes. Even when subject to open access mandates like that of UKRI, use for machine learning is being hampered by some publishers increasingly ignoring open access (CC BY) mandates. Instead some of them are encouraging academics to sign more restrictive licences that don’t allow commercial use of their works or the creation of adaptations of the original. However, in this process what academics are in fact doing is signing away these rights exclusively to the publisher, reinforcing their role as gatekeepers to publicly funded research. In turn, it is surely no coincidence that many of the same publishers are in turn signing AI deals with large technology firms.
If the government was serious about supporting maximum impact from its investment in research, alongside creating a more flexible copyright regime, what it should be doing is following the lead of many European countries in passing legislation that ensures that outputs from publicly funded research can be made immediately available in open access repositories irrespective of where first published – something called a secondary publishing right or in Scandinavia sometimes called a “parallel publishing” right.
Better Data Laws Will Support Growth
As pointed out above, some publishers continue to act as gatekeepers of publicly funded science, despite the fact that in a research-intensive digital economy like that of the UK, ensuring that the “data” that results from taxpayer funded science is available to everyone has such clear social and economic benefits.
Among the many questions raised by the rapidly moving geo-political situation, we should not underestimate the benefits to the UK research sector’s that could arise from a strong measure of autonomy in both the development and application of AI. The potential consequences both for the sector, and for our wider competitiveness make this not a theoretical aspiration but an immediate strategic necessity.
Without steps to build this autonomy, universities and researchers risk becoming dependent consumers of opaque technologies, constrained data regimes, and foreign laws – undermining the national interest, innovation, and economic impact. With it, they can remain engines of innovation and learning, trusted partners to industry and the public sector, and stewards of publicly funded knowledge.
The choice is stark: invest now in the skills, infrastructure, governance, and data freedoms needed for AI sovereignty, or accept a gradual erosion of the UK’s research capacity in an increasingly competitive and fragmented global landscape. The window to act is open.
27 March 2026
