This essay is the author's independent view and does not represent the position of the University of Bradford.
Balanced in whose interests?
On 27 October 2026, Jisc published a sector-wide statement calling for greater governance, transparency and cost control in the adoption of artificial intelligence across UK higher and further education (Jisc, 2026a). Supported by organisations including Universities UK, GuildHE and the Association of Colleges, the statement presents an apparently balanced and entirely reasonable position.
AI offers considerable opportunities, but institutions must protect themselves against escalating costs, supplier dependency, unreliable services and the ethical and educational problems associated with its deployment.
On the face of it, there is little to disagree with. I would be rather concerned if universities were entering expensive, long-term AI agreements without asking some fairly elementary questions about what they were purchasing.
Nevertheless, I am not entirely convinced by this apparent balance.
My concern is less about what Jisc advocates than the institutional assumptions upon which its advice rests.
Jisc is deeply embedded within the technological infrastructure of UK education. It provides services, negotiates licensing agreements, advises on procurement and helps institutions develop their digital strategies. Its own activities, and consequently its organisational interests, are closely associated with the institutions it serves.
This relationship is hardly surprising. Jisc exists to support the sector, and much of its work is both necessary and valuable. But it raises a question that seems to receive rather less attention than it deserves.
Can an organisation whose activities are so closely intertwined with existing institutional arrangements offer an entirely balanced assessment of a technology capable of fundamentally changing those arrangements?
I suspect the answer is more complicated than the familiar language of responsible innovation might suggest.
What exactly is being balanced?
The word balanced is reassuring. It suggests that competing perspectives have been considered, risks weighed against benefits and a sensible position reached somewhere between uncritical enthusiasm and outright resistance.
Yet balance depends upon what is being weighed, whose interests are represented and what has been excluded from consideration.
Jisc's October statement concentrates upon contractual transparency, pricing, supplier accountability, interoperability and institutional control (Jisc, 2026a). These are legitimate concerns, particularly as AI providers introduce increasingly complex licensing and consumption-based pricing arrangements.
However, the interests being considered are predominantly those of institutions as purchasers, administrators and governors of technology.
Where does the individual academic, researcher, technician or professional services employee fit into this conception of balance?
An institution may regard a centrally approved AI platform as the sensible option because it offers predictable expenditure, established security arrangements and convenient integration with existing systems.
Its employees, meanwhile, may find that same platform unsuitable for particular intellectual or professional activities.
There is no great mystery here. What constitutes a sensible purchasing decision for an IT department may be quite different from what constitutes a useful working tool for an academic.
The difficulty arises when institutional convenience becomes synonymous with professional suitability, and the requirements of central administration are treated as though they necessarily represent the interests of those undertaking the work.
Responsible innovation involves more than managing organisational risk. It should also involve examining whether existing practices remain appropriate when new technological possibilities emerge.
I am less certain that this second consideration receives comparable attention.
Innovation within the boundaries of what already exists
Jisc's Leading AI in Colleges: A Strategic Framework, published in November 2025 by its director of AI, Michael Webb, provides an instructive example.
The framework is specifically intended for further education colleges, where safeguarding responsibilities, financial constraints and staff development requirements differ in important respects from university research environments. It would therefore be inappropriate to treat its recommendations as evidence of an established policy governing academic freedom in higher education.
Nevertheless, it offers an insight into Jisc's general approach to institutional AI adoption.
The framework organises its recommendations around three areas: skills and culture, technology, and governance. It encourages colleges to identify practical problems, develop staff expertise, protect information and establish appropriate oversight (Webb, 2025).
Much of this is sensible.
I particularly welcome its recommendation that institutions identify problems before selecting technologies. Anyone who has spent a reasonable amount of time working in education will recognise the tendency to acquire systems and subsequently discover reasons why they were indispensable.
What interests me is the emphasis placed upon existing technological arrangements.
Colleges are encouraged to examine AI capabilities within software they already license, make use of established suppliers and develop customised applications through their existing cloud platforms.
For institutions heavily invested in Microsoft, this may involve Copilot and PowerApps. Those operating within Google's ecosystem are directed towards corresponding Google services.
The attractions are obvious. Existing contracts, familiar systems and established security arrangements can reduce both costs and implementation difficulties.
But such advantages also introduce a potential bias.
David's (1985) account of technological path dependence is relevant here. Earlier technological choices influence subsequent decisions, sometimes encouraging particular systems to persist because of accumulated investment, compatibility requirements and the costs associated with changing direction.
Universities are not approaching AI with a blank sheet of paper. They already possess substantial technological infrastructures, established contracts and administrative procedures.
It is hardly surprising that they might prefer AI technologies that fit neatly within those arrangements.
The problem is that compatibility with what an institution already possesses tells us relatively little about whether a particular technology is the most appropriate choice for the work being undertaken.
The easiest AI system for a university to administer is not necessarily the best AI system for its staff to use.
This distinction becomes particularly important when considering the diversity of activities carried out within universities.
Task–technology fit: One institution, many different needs
Goodhue and Thompson's (1995) model of task–technology fit provides a useful framework for examining this problem.
Their central proposition is that the contribution of information technology to individual performance depends partly upon how well its capabilities correspond to the requirements of the tasks being undertaken.
The model was not developed specifically for generative AI, but its underlying logic seems particularly relevant.
Consider three employees working within the same university.
A departmental administrator uses AI to summarise meetings, organise correspondence and locate information distributed across institutional documents.
A psychology lecturer wishes to explore competing theoretical accounts of cognition, assess the limitations of published research and develop new conceptual arguments.
A technical specialist wants to construct an interactive statistical teaching application, develop computational simulations or troubleshoot a piece of research software.
Each may benefit considerably from AI, but their requirements are rather different.
For the administrator, integration with Microsoft 365 may be especially valuable. Access to existing documents, calendars and organisational information can make an integrated enterprise platform highly effective.
For the lecturer, the quality of reasoning, ability to evaluate conflicting evidence and reliability of source attribution may be more important.
For the technician, coding performance, mathematical competence and the ability to develop and test working applications may determine whether a particular tool is useful.
There is no obvious reason why the same AI platform should necessarily be optimal for all three.
Yet institutional procurement often operates through economies of scale, standardisation and existing supplier relationships.
This introduces a potentially important conflict between organisational efficiency and individual task–technology fit.
It is also a conflict that could be investigated empirically.
Do employees using centrally selected AI platforms perform particular tasks as effectively as those permitted to select from several appropriately governed alternatives?
How frequently do staff find that approved systems cannot adequately support their work?
What are the consequences for productivity, innovation and professional satisfaction?
These seem worthwhile questions for institutions claiming to pursue evidence-based AI adoption.
After all, selecting technology according to the requirements of the task would appear to be a fairly elementary principle.
Copilot and the problem of presumed suitability
Microsoft Copilot provides a useful illustration, although the argument is not primarily about Microsoft.
Jisc includes Copilot within its guidance and offers training in Microsoft 365 Copilot for professional use. It also identifies ChatGPT and Gemini as general-purpose alternatives, so it would be inaccurate to suggest that Jisc advocates Copilot exclusively (Webb, 2025).
Nevertheless, Copilot is likely to be attractive to institutions already operating extensive Microsoft environments.
That attraction has legitimate technical foundations. Microsoft 365 Copilot can work within established organisational permissions and integrate with applications used routinely by university staff.
However, enterprise integration and model reliability are different properties.
A platform may offer excellent access to institutional information while being less effective for particular forms of reasoning, coding or research. Conversely, a system that performs well on specialised intellectual tasks may offer inadequate safeguards for confidential organisational information.
Neither consideration should be ignored.
The problem arises when satisfactory performance on one dimension is taken as evidence of general suitability.
Microsoft 365 Copilot should also be distinguished from other products using the Copilot name, including development tools and services that may employ different models or operating arrangements. Comparisons between AI systems must account for those differences.
I would not suggest that ChatGPT, Claude or any other platform should automatically be preferred. Their capabilities are continually changing, and all remain susceptible to error.
The point is that suitability must be demonstrated.
A university that adopts an enterprise AI platform for administrative purposes has not thereby established that the same platform is the most appropriate tool for every academic or technical activity.
That conclusion requires evidence relating to the activities concerned.
Without it, there is a danger that procurement becomes a substitute for evaluation.
The psychology of institutional continuity
The tendency for organisations to reproduce familiar structures is hardly new.
DiMaggio and Powell (1983), in their influential account of institutional isomorphism, examined why organisations operating within the same field often become increasingly similar. They identified coercive, mimetic and normative pressures that encourage conformity, particularly under conditions of uncertainty.
Their account seems particularly relevant to higher education.
One university establishes an AI governance committee. Others follow. Policies are written, approved platforms identified, training programmes developed and compliance procedures introduced.
Before long, a recognisable model of responsible AI adoption begins to emerge across the sector.
Some of this coordination is beneficial. Institutions should not need to independently rediscover solutions to identical problems involving information security, procurement and data protection.
Nevertheless, the appearance of a coordinated response should not be confused with evidence that the resulting arrangements are necessarily optimal.
Universities may adopt particular governance arrangements partly because they resemble what other respectable institutions are doing.
This is where organisations such as Jisc acquire considerable influence. Their recommendations can help define acceptable practice, creating a framework within which subsequent decisions are made.
Yet there is a further psychological consideration.
Organisations develop established ways of interpreting problems. These reflect their history, responsibilities, expertise and relationships with other organisations.
A university IT department is likely to understand AI partly through security, integration, software procurement and infrastructure.
An academic may approach the same technology through intellectual exploration, research methodology or teaching.
Neither perspective is inherently mistaken. They simply attend to different aspects of the problem.
Difficulties arise when one perspective acquires sufficient institutional authority to define the problem for everyone else.
Jisc's established expertise lies substantially in supporting the digital infrastructure and technological needs of educational institutions.
Consequently, its interpretation of AI adoption will understandably place considerable emphasis upon infrastructure, licensing, governance and institutional capability.
That emphasis may be entirely sincere while remaining incomplete.
An organisation can be very good at solving the problems it recognises while being less inclined to recognise problems that challenge the necessity of its existing solutions.
This is where I think institutional self-preservation becomes a useful hypothesis.
It need not involve deliberate resistance, dishonesty or some coordinated attempt to protect established interests. It may emerge naturally from the way organisations understand their responsibilities and define what constitutes a reasonable response to change.
Technological innovation can consequently be welcomed, encouraged and generously funded, provided it remains intelligible within established institutional arrangements.
There is something rather familiar about that.
Governance and professional autonomy
Jisc's framework gives considerable attention to governance, including institutional policies, procurement, monitoring and senior management accountability (Webb, 2025).
These recommendations reflect genuine responsibilities. Universities and colleges cannot ignore data protection, contractual obligations, safeguarding or information security.
Jisc also recognises that different disciplines require different approaches and cautions against unnecessary standalone AI policies. Its guidance is therefore more flexible than a simple argument about bureaucratic control might suggest.
Nevertheless, there remains an important distinction between protecting institutional information and prescribing the intellectual tools through which professionals undertake their work.
Preventing staff from entering identifiable student information into an unsecured AI service is entirely reasonable.
Preventing an academic from using an alternative AI platform to develop a conceptual model, where no confidential information is involved, is a rather different matter.
The first restriction protects information.
The second potentially constrains professional judgement.
Universities have traditionally attached considerable importance to academic freedom and intellectual independence. These principles do not remove legitimate professional responsibilities, but they do establish an expectation that academics should exercise informed judgement in selecting methods appropriate to their work.
That expectation ought to extend, at least in principle, to AI.
There is an additional irony.
AI has considerable potential to broaden intellectual participation, particularly among individuals who may struggle with the conventional language, confidence or social expectations associated with academic discourse.
It can help people develop and articulate ideas that might otherwise remain unexplored.
This does not transform an uninformed individual into an expert, nor does it remove the necessity for critical evaluation. But it can provide a means of participating in intellectual activities previously mediated through established institutional and professional channels.
Universities should be particularly interested in such possibilities.
It would be unfortunate if access to those possibilities became unnecessarily constrained by historical software purchasing decisions.
A proportionate approach would distinguish between the sensitivity of information, the consequences of error and the professional purpose of the activity.
Exploratory work involving publicly available information need not be governed in precisely the same way as decisions concerning student welfare, confidential research data or institutional finances.
Such distinctions would permit professional experimentation while preserving appropriate safeguards.
I would regard that as a rather more convincing expression of balance.
Jisc's organisational interests
Jisc occupies an interesting position within this emerging landscape.
It advises institutions on technological adoption while also providing services through which that adoption is implemented.
Its activities include collective licensing, technological infrastructure, consultancy and professional development.
In commentary accompanying its 2026 higher education leadership survey, Jisc identifies expanding AI licensing arrangements and institutional consultancy as important areas of work. It also acknowledges that its recognition as a leader in AI is lower than it would like (Jisc, 2026b).
There is nothing unusual about an organisation considering its future relevance.
Nevertheless, Jisc's advisory and service-provision roles deserve examination when evaluating its recommendations.
A growing emphasis upon AI governance may increase demand for precisely the expertise and services that organisations such as Jisc provide.
That does not establish that the demand is artificial or that its recommendations are motivated by financial self-interest.
It establishes a potential organisational incentive.
Once an organisation becomes recognised as a source of expertise in managing a particular problem, its services, professional assumptions and institutional relationships may develop around the continued importance of that problem.
Governance can acquire its own momentum.
Risk assessments require review. Policies require monitoring. Approved applications require evaluation. Evaluation produces recommendations that may generate further procedures, training and consultancy.
Some of this activity will be essential.
But who determines when the administrative apparatus has become disproportionate to the risks it was intended to manage?
And who evaluates the professional opportunities lost through compliance with that apparatus?
These are familiar questions in higher education, where substantial resources are already devoted to measuring and assuring activities whose underlying educational value is not always easy to establish.
AI governance could become another example.
The concern is not that organisations deliberately manufacture unnecessary work. It is that established organisational practices can continue expanding without anyone seriously questioning whether they remain proportionate or necessary.
The strongest argument in Jisc's defence
Jisc's October 2026 statement contains an important counterargument to this critique.
It explicitly calls for greater supplier transparency, interoperability and the ability to move between large language models without unnecessary disruption or additional costs (Jisc, 2026a).
These demands are significant.
If universities become dependent upon a handful of commercial AI providers, they may face escalating costs, restrictive contracts and limited control over the technologies upon which their activities depend.
Collective bargaining can help address these problems.
Jisc's negotiating position may provide smaller institutions with commercial influence they would otherwise lack. Shared expertise can reduce duplicated effort, while common standards may help preserve flexibility and protect institutions from unfavourable contractual arrangements.
Far from preserving technological dependency, Jisc could help universities escape it.
This possibility deserves serious consideration.
It would be naïve to assume that removing established educational intermediaries would necessarily democratise access to AI. We might simply replace dependence upon university institutions with dependence upon a small number of powerful technology companies.
That would hardly represent an unqualified improvement.
Nevertheless, an important distinction remains.
Reducing the costs and restrictions associated with existing technological arrangements is not necessarily the same as examining whether those arrangements should continue.
Jisc's position is strongest when it challenges commercial dependency.
It is less clear how far its approach extends towards questioning the organisational structures through which that dependency has developed.
The two objectives are compatible, but they are not identical.
A genuinely transformative approach would need to address both.
When efficiency is mistaken for transformation
This brings me to what I consider the most important issue.
Much of the discussion surrounding institutional AI adoption assumes that universities will continue performing broadly recognisable functions, albeit with increasingly sophisticated technology.
The problem becomes one of introducing AI into existing workflows, improving efficiency and reducing costs.
But what if AI changes the need for some of those workflows?
If substantial elements of administration can be eliminated, why should institutions concentrate solely upon making them more efficient?
If particular services can be delivered through simpler arrangements, why preserve the structures through which they have traditionally been provided?
If individuals can undertake increasingly sophisticated intellectual activities without relying upon established intermediaries, how should universities respond?
These questions do not imply that universities are becoming unnecessary.
Higher education serves important intellectual, social and research functions that cannot simply be reduced to the production and processing of information.
However, that does not establish that every existing administrative process, professional role or organisational arrangement remains indispensable.
Nor should the value of an activity be inferred from the fact that an institution has historically employed people to undertake it.
The distinction between institutional efficiency and institutional transformation matters.
An organisation can become extraordinarily efficient at doing something that no longer needs to be done.
That would be progress of a rather peculiar kind.
This is where Jisc has an opportunity to contribute something more substantial than guidance on responsible procurement.
Its sector-wide experience could help universities examine which existing arrangements are worth preserving, which require fundamental redesign and which may no longer serve a useful purpose.
Such an exercise would inevitably challenge established interests, including potentially those associated with Jisc's own services.
But a willingness to contemplate those consequences would be a more convincing demonstration of independence than any number of statements about balanced technological adoption.
Towards a more convincing balance
A genuinely balanced approach to AI governance would recognise that institutional interests are only one consideration.
Student welfare, information security, financial sustainability and academic integrity are clearly important.
So too are professional autonomy, intellectual experimentation, accessibility and technological diversity.
These interests may conflict, but that is precisely why they require consideration rather than being subsumed beneath a single institutional definition of responsible practice.
I would suggest several requirements.
Institutions should distinguish restrictions on sensitive information from restrictions on general-purpose intellectual tools. Where an activity involves no confidential data, limitations upon professional choice should require a clear and proportionate justification.
AI platforms should be evaluated according to the tasks for which they are intended, using evidence of performance, reliability, accessibility, security and cost.
Staff should have meaningful opportunities to challenge approved-tool decisions and propose alternatives. Their expertise should contribute to institutional decisions rather than merely being expected to accommodate them.
Policies should also be reviewed frequently enough to reflect technological developments. A platform considered unsuitable today may become highly effective tomorrow, while an established product may be overtaken by competitors.
Finally, organisations advising the sector should be transparent about their commercial relationships and institutional interests, including the assumptions underlying their recommendations.
None of this requires abandoning governance.
Indeed, it might make governance considerably more effective.
The objective should be to support professional work while managing demonstrable risks, rather than treating administrative manageability as the principal measure of technological suitability.
Conclusion: The managed revolution
Jisc is not an opponent of artificial intelligence. Its published activities demonstrate a clear commitment to supporting adoption, and many of its concerns are entirely justified.
The difficulty is that an organisation can sincerely encourage innovation while understanding that innovation principally through the responsibilities, services and institutional relationships it already possesses.
This is where the psychology of institutional continuity becomes important.
Organisations do not necessarily preserve existing arrangements because their members consciously fear change. They may do so because established structures shape which problems they recognise, which solutions appear reasonable and what they understand successful innovation to mean.
An institution selects a familiar AI platform because it integrates with existing systems. It develops governance policies because other institutions have done so. It establishes additional monitoring because the policies require assurance.
Every decision can be justified.
Yet the cumulative result may be a considerable extension of institutional control over a technology whose potential lies partly in enabling people to work beyond established organisational boundaries.
This is what troubles me about Jisc's apparently balanced position.
It is balanced in its consideration of institutional risks. I am less convinced that it gives comparable attention to professional autonomy or the possibility that AI may require more fundamental changes to the institutions themselves.
Perhaps Jisc's greatest contribution to higher education would be to help universities identify not only where AI can improve existing practices, but where those practices are no longer necessary.
That might sometimes require questioning arrangements through which Jisc itself currently operates.
I would regard such willingness as a meaningful test of institutional independence.
After all, if a technological revolution must first be made compatible with everything it might otherwise change, one is entitled to wonder how much of a revolution remains.
References
David, P. A. (1985). Clio and the economics of QWERTY. The American Economic Review, 75(2), 332–337. https://www.jstor.org/stable/1805621
DiMaggio, P. J., & Powell, W. W. (1983). The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147–160. https://doi.org/10.2307/2095101
Goodhue, D. L., & Thompson, R. L. (1995). Task–technology fit and individual performance. MIS Quarterly, 19(2), 213–236. https://doi.org/10.2307/249689
Jisc. (2026a, October 27). Education and research sector calls for governance and cost control for AI. https://www.jisc.ac.uk/news/all/education-and-research-sector-calls-for-governance-and-cost-control-for-AI
Jisc. (2026b, May 26). What we learned from our tenth higher education sector leadership survey. https://www.jisc.ac.uk/blog/what-we-learned-from-our-tenth-higher-education-sector-leadership-survey
Webb, M. (2025, November 17). Leading AI in colleges: A strategic framework. Jisc. https://www.jisc.ac.uk/guides/leading-ai-in-colleges-a-strategic-framework