Higher education governance under AI and geopolitical disruption: a socio-technical framework for administrative artificial intelligence

Synthesis: Pandey and Davies argue that administrative AI — chatbots, predictive analytics and automated screening applied to recruitment, admissions and student services — has reached the administrative core of universities at a moment when geopolitical disruption is making international enrollment volatile. Their conceptual framework names five drivers of adoption, reads them through socio-technical systems theory, and works the reading against institutional cases and documented episodes of failure. The central claim is that administrative AI poses a governance question rather than a technical one: whether it produces resilience or dependency turns on procurement terms, contestability, participation and what gets audited.

Conceptual paper: the authors present a framework and three propositions, not data. No systems were evaluated, no institutions surveyed and no model tested. Every figure below is attributed to the study or agency that produced it, and the governance conditions are implications derived from the analysis rather than findings from institutional evidence.

What the framework argues

1. Administrative AI is a defined object, not a general label. The authors mean AI applied to administrative services rather than to teaching, learning or research, concentrated on front-office work: recruitment and admissions, enrollment, student records, advising, housing, welfare and careers.
2. The boundary with teaching is analytical, and the paper says so. Administrative systems reach academic work by shaping cohort composition, sending risk flags to academic desks as requests to intervene, and fixing the data environment through procurement.
3. Three propositions carry the socio-technical reading. Joint optimization holds that altering technology without redesigning roles, skills and accountability impairs performance. The politics of artifacts holds that assumptions enter through training data, categories, optimization targets and thresholds. Globally shaped infrastructure holds that whatever makes the technology feasible also makes the institution dependent.
4. Five drivers explain adoption, each with a mechanism, actors and a characteristic risk. They are infrastructural readiness, competitive positioning under enrollment volatility, regulatory and mobility mandates, commercial platform ecosystems, and Sustainability legitimacy. The first and fourth enable and steer adoption; the second, third and fifth motivate it.
5. Combinations, not single drivers, predict the pathway. Mandate without infrastructure yields symbolic adoption; infrastructure and vendor push without a mandate yield product roadmaps; volatility under acute pressure yields systems that are brittle when resilience is most needed.
6. Governance conditions and leadership capabilities are the payoff. Conditions are attached to the actor able to act on them, and four capabilities are named: contractual literacy, interrogating the objective, restraint under pressure and standing participation.

Why an efficiency framing misses the governance question

If administrative AI is classified as routine efficiency, the authors argue, three things disappear: that it moves work around, that it converts values into procedures, and that it ties a university to data infrastructures governed elsewhere.

The discretion argument is borrowed from public administration and applied where it is rarely asked. Bovens and Zouridis traced how information technology moved discretion from frontline staff to system designers, producing screen-level and then system-level bureaucracies. Their question was constitutional: where has discretion gone, and who answers for it now? The authors put it to admissions, progression and student support, where a threshold stands in for a judgment a person used to make.

Surveillance is defined in Lyon's technical sense as focused, systematic, routine attention to personal details in order to influence, manage or control someone. Much administrative AI qualifies regardless of intent, and whether it is objectionable depends on proportionality and on whether anyone can challenge it.

The theoretical contribution is a coupling the authors claim neither tradition predicts alone. Volatility supplies the reason to automate and shortens the time available to absorb the system. The technology bought to absorb the shock is then procured from outside the jurisdiction whose policy produced it. Exposure and remedy run through one channel, so the institution most exposed to a mobility shock is least able to govern the system it buys to manage that shock.

Where the administration and learning boundary sits

Administrative AI covers conversational agents answering applicant inquiries, predictive analytics forecasting enrollment or flagging students at risk, and automated or assisted screening in recruitment, admissions and progression. Teaching, research and the back office sit outside the scope.

A model settling which applicants look promising also settles the composition of the cohort a curriculum is designed for. Risk flags arrive on academic desks as requests to intervene, changing what advising asks of teaching staff. Procurement carries furthest, because the platform holding student records sets the data environment in which research ethics, institutional research and learning analytics all operate. Administrative AI reaches teaching through consequences, not through any classroom tool.

A single risk-based instrument could govern both halves, as Temper et al.'s teaching-life-cycle framework would, so the boundary is an analytical convenience rather than a firewall.

The five drivers, and the paths they send institutions down

Infrastructural readiness works by enablement. Cloud storage, fast processors and reliable data pipelines lower the threshold at which adoption becomes practicable. Dependency arrives packaged with feasibility.

Competitive positioning works through advantage-seeking and risk anticipation in markets exposed to migration policy. The illustration is Canada's January 2024 cap on new international study permits, intended to cut admissions by 35 percent against 2023. Just under 268,000 permits were granted that year, a fall of 48 percent that left the government roughly 100,000 permits short of its own target, and program suspensions followed, on the ICEF Monitor account the authors cite. For institutions dependent on international enrollment, analytics stops being an instrument of advantage and becomes an early-warning system. The risk is that yield, conversion and ranking position quietly define which applicants count.

Regulatory and mobility mandates arrive through two channels the literature keeps apart. Education policy constrains as it pushes, through mandates, grant conditions and sector strategies. Migration compliance pushes without constraining, through visa regulation, attendance monitoring and right-to-study checks. Intrusive monitoring is then imposed on institutions rather than chosen by them, and some of the most intrusive data practices in higher education are not chosen by universities at all.

Commercial ecosystems work on the supply side. Universities rarely build these systems, and regional hubs build quickly with no legacy platforms to work around. Administrative AI arrives bundled inside customer relationship management suites, enrollment platforms and student information systems already in use, and vendors settle what an AI-enabled admissions office should look like before local discussion begins. Lock-in, lost data sovereignty and eroding autonomy follow, hardest on institutions with the least in-house legal and technical capacity.

Sustainability legitimacy works through audiences that judge institutions against development agendas. Leadership teams invoke SDG 4 on quality education and SDG 9 on infrastructure when arguing for investment, and the risk is symbolic alignment: deployment reported as contribution while distributional effects go unmeasured.

Mandate without infrastructure produces stalled or symbolic adoption. Infrastructure and vendor push without a mandate produce adoption defined by product roadmaps, the pattern the authors find in many institutions with limited resources. Volatility under acute pressure produces systems bought but never absorbed. Institutional isomorphism explains why the same tools arrive in institutions with little else in common, and leaves open why they do not converge on the same exposure.

What the counter-cases add

A literature written only about adoption acquires survivor bias, so the paper turns to failure. Five episodes are tabulated. At St George's Hospital Medical School, a screening program reproduced its staff's discrimination, as the Commission for Racial Equality found in 1988. Ofqual withdrew its 2020 standardization of teacher-assessed A-level grades within days, and the Home Office suspended its visa streaming tool in 2020 before a judicial review. DUO halted fraud-risk profiling with a ministerial apology and compensation, and India's National Scholarship Portal shows biometric failures falling hardest on manual workers.

In each case a system took a high-stakes decision whose workings stayed invisible to the people affected, each was calibrated against a historical record, and none offered an effective route of appeal. Four of the five involved students at the boundary of an institution or a state — the population volatility makes both larger and more vulnerable. The correction came from outside every time.

Mapped onto the drivers, the cases sort by mechanism. Ofqual and the visa tool are mandate cases, imposed at speed with no organizational absorption behind them. DUO and the scholarship portal combine mandate and infrastructure. St George's is the outlier, a competitive-positioning case with no external mandate that failed anyway. Failure follows wherever a driver supplies motive or requirement without the redesign joint optimization demands.

What this means for practice

• Institutional leaders. Require an equity impact assessment before any AI enters an admissions, recruitment or progression decision, testing outputs against biases already in the training data, and publish a register of algorithmic systems used in decisions affecting students.
• Procurement. Treat the contract as the governance instrument. Data sovereignty, audit rights, exit and portability terms decide whether the institution bought resilience or dependency.
• Sector bodies and policymakers. Release model procurement terms no single institution could negotiate alone, fund shared regional and Open Source infrastructure, and extend accountability across the mobility pipeline so appeal rights do not depend on whether a university or a state agency decided.
• Ranking and reporting bodies. Require any SDG-linked claim about administrative AI to report distributional outcomes by the groups SDG 10 names, and decline deployment counts in their place.
• Researchers. The two gaps forcing hedging throughout — administrative AI in lower-income institutions, and systematic documentation of counter-cases — are the comparative work the framework is offered to support.

Limitations

• This is a conceptual analysis and an interpretive review, not a systematic one; the authors state that it stands or falls on coherence, and no data were collected to test it.
• The literature base is mostly Anglophone and skewed toward well-resourced institutions, so claims about lower-resource systems and emerging hubs are explicitly provisional.
• The three propositions are stated in testable form but remain untested; no causal claim about adoption, resilience or equity is established.
• Counter-case selection rested on three stated conditions, and remote proctoring is retained with a scope caveat because it sits at the edge of the paper's own definition.
• The governance conditions, leadership capabilities and SDG alignments are implications derived from the analysis rather than findings from institutional data.

Citation

Pandey, N., & Davies, D. (2026). Higher education governance under AI and geopolitical disruption: a socio-technical framework for administrative artificial intelligence. Frontiers in Education, 11, 1997227.