The new bargaining frontier: AI and academic labour

Higher education has spent the past several years asking what artificial intelligence will do to teaching. Will students use it to write assignments? Should professors allow it in the classroom? Can it improve feedback, personalise learning, or accelerate research?

Universities have established task forces, issued guidelines and purchased new platforms, while faculty have been asked to rethink assessment almost as quickly as the technology evolves.

Yet beneath these highly visible debates lies another question that may prove more consequential for the future of higher education: What happens when AI begins to perform work that universities have historically employed people to do?

At that point, AI ceases to be merely an educational technology and becomes a labour technology.

Once AI enters the employment relationship, questions arise that an information technology office alone cannot resolve. Who decides whether AI may perform work previously assigned to faculty? Can AI change staffing requirements?

Who owns academic materials used to develop AI systems? Can algorithmic analysis influence faculty evaluations? If AI makes academic work faster, who benefits from the time saved? If supervising AI creates new work, where does that labour appear in workload calculations?

These questions are no longer hypothetical. They are increasingly appearing in collective agreements, memoranda of understanding and union policies.

AI has already reached the bargaining table

At the City University of New York, the 2023-27 agreement between CUNY and the Professional Staff Congress requires that the instructor of record for every course hold an instructional staff title. The union describes this provision as a safeguard against outsourcing teaching functions to artificial intelligence.

The agreement also establishes an educational technology labour-management committee whose discussions include artificial intelligence, technology training, instructional design and the impact of online teaching modalities on the terms and conditions of employment.

The relevant provisions are set forth in the PSC-CUNY 2023-27 agreement. The union has since discussed the implications for AI and academic employment in its coverage of protecting CUNY jobs from AI.

In June 2024, Miami University in Ohio and the Faculty Alliance of Miami, AAUP-AFT, signed an artificial intelligence memorandum of understanding. The agreement provides for labour-management discussions on AI’s effects on bargaining unit faculty and allows both parties to propose guidelines for specific AI programmes and tools.

The Miami case is particularly instructive because it also shows why the legal dimensions of AI bargaining should not be overstated.

The union argued that AI affected terms and conditions of employment and therefore required bargaining, while the university contested the proposition that AI itself was automatically a mandatory subject of bargaining. A subsequent AI task-force MOU documented the ongoing process.

The lesson is important. No universal rule makes ‘AI’ a mandatory bargaining subject across higher education systems. Labour law, existing contracts and institutional governance arrangements vary by jurisdiction. However, when AI affects workload, evaluation, intellectual property, job security, or other terms of employment, it can become a legitimate bargaining issue.

At the Community Colleges of Spokane in Washington state, the 2025-28 Master Contract explicitly recognises AI as a rapidly evolving issue and requires ongoing discussion between the colleges and the faculty association.

Unresolved AI-related questions can be referred to a joint executive committee. Reporting by the National Education Association shows how such provisions can matter when new AI functions are proposed for institutional technology systems.

Outside the United States, academic unions and associations are moving in the same direction.

In November 2025, the Canadian Association of University Teachers (CAUT) approved a policy on generative AI to help academic staff associations develop collective agreement language and institutional governance. The policy addresses job security, workload, intellectual property, privacy, collegial governance and protections for contract academic staff.

In the United Kingdom, the University and College Union’s principles for the use of artificial intelligence state that the introduction of AI technologies that affect staff and working conditions should be subject to consultation and agreement through recognised collective bargaining processes before implementation. UCU also calls for the protection of staff-created intellectual property and greater transparency in AI procurement.

At the global level, Education International convened more than 200 union leaders, educators and experts in Brussels in December 2025 for its first global conference focused specifically on AI in education, with additional participants joining remotely.

Its conference report shows that AI and the future of educational work are now part of the international trade-union agenda.

When AI performs work, whose work is it considered?

The first labour question is deceptively simple: Can AI perform work traditionally done by faculty or other university employees?

Universities have long grappled with questions surrounding subcontracting and outsourcing. AI introduces a different version of the same structural problem because work does not necessarily move to another employee or an external contractor. Some portion of it may instead move into a computational system.

Consider academic advising. AI systems can answer routine questions about course requirements, schedules and degree progression. Generative systems can produce instructional materials, provide preliminary feedback, summarise student enquiries and assist with routine administrative writing.

The labour question begins before anyone loses a job: it begins when the boundary of the job shifts. A university might retain professors while redistributing work among them – faculty design the course, AI generates supporting materials and handles routine questions, and professional staff intervene only when cases exceed the system's capacity. This is not necessarily a replacement; it is a reconfiguration of academic labour.

The concern is especially relevant to contingent and contract academic staff, whose positions may already lack the employment protections afforded to permanent faculty.

The CAUT policy highlights the vulnerability of contract academic staff, while the AAUP's 2025 report on artificial intelligence and the academic professions identifies concerns about job security, wages, professional autonomy and increased reliance on contingent academic labour.

The union question should therefore not be framed so simply: Will AI replace professors?

A more useful question is: Which parts of academic work become easier to remove, reorganise, centralise or devalue once institutions discover that machines can perform portions of cognitive labour?

The labour that AI does not eliminate

Another problem lies within the promise of automation.

AI also creates work.

A generated lecture outline must be reviewed. A literature synthesis must be verified. Citations may need to be traced to their original sources. Outputs must be assessed for errors, fabricated information, bias and inappropriate assumptions.

Faculty must determine which student information can safely be entered into commercial systems. Assignments and examinations may need to be redesigned as generative AI changes what students can produce independently.

I would describe this emerging category as AI verification labour.

A professor may save time by generating a preliminary teaching resource but then spend considerable time verifying its reliability. An instructor may use AI-assisted feedback while still retaining professional responsibility for ensuring that the recommendations are pedagogically appropriate.

A department may adopt AI to reduce workload, while faculty simultaneously spend additional hours on training, rewriting policies and redesigning assessments.

The AAUP's 2025 report explicitly identifies workload intensification as a concern and notes that professional development and the implementation of new technologies can increase workload.

The American Federation of Teachers' principles for AI in higher education similarly argue that AI should not increase workload without corresponding compensation or a workload adjustment.

This raises a practical collective-bargaining question: Where does AI verification labour appear in the workload model?

If an institution expects faculty to redesign assessments because of generative AI, is that considered academic work? If mandatory AI training takes several days each year, is it counted as part of the workload? If professors remain professionally responsible for AI-supported outputs, does the institution provide the time needed to fulfil that responsibility?

Universities should be careful not to count only the minutes AI saves while leaving the hours required to govern it uncounted.

Who receives the productivity dividend?

Suppose AI genuinely reduces the time spent on repetitive academic and administrative work. Preparing routine teaching materials becomes faster, literature discovery improves, correspondence takes minutes rather than hours, and data analysis becomes easier.

Who owns the time AI saves? If an academic previously needed five hours to complete a set of tasks and can now do so in two, those three hours represent a productivity gain.

They could support more research, mentoring and intellectual engagement with students – or justify another course, more advisees, larger classes, faster response expectations and higher publication targets.

I would call this the productivity dividend of augmented academic labour.

Again, I use the phrase as a conceptual proposition rather than as an established term in labour law. The underlying distributive question, however, is already present in union policy.

The AFT's 2026 resolution on artificial intelligence argues that workers should share fairly in the economic gains from technological change and calls for bargaining rights over the evaluation, procurement, implementation, monitoring and workplace effects of AI.

That is an advocacy position, not a universal legal entitlement. But it raises a question that universities will eventually struggle to avoid.

If AI enables one academic to produce what previously required significantly more human labour, who receives the value created by that additional capacity? The institution? The technology company? Students? The faculty member? Or all of them?

One of the most important AI negotiations in higher education may therefore focus not on banning AI or protecting a particular academic task but on how the gains from human-machine augmentation are distributed.

Can a university reproduce a professor's labour?

Intellectual property makes the labour problem even more complicated. Universities possess enormous amounts of faculty-generated material: recorded lectures, slides, syllabi, assessment banks, course designs, written feedback and research outputs.

There is an important difference between keeping a recording of a professor's lecture and using accumulated lectures and course materials to configure a system capable of generating new teaching content derived from those materials.

The CAUT policy on generative AI states that an academic staff member's image, voice and professional or academic works should not be reproduced, distributed, or monetised by an institution without authorisation.

UCU's AI principles likewise state that staff-generated copyright material should not be used to train AI systems without consent and call for fair treatment when staff intellectual property is commercially exploited.

Imagine a professor who has spent 20 years developing a distinctive course. The university has recordings, slides, assignments and years worth of instructional materials.

If those materials are used to develop an AI system that reproduces significant elements of the professor's teaching, what exactly does the university own? The files? The course? Or a computational approximation of a portion of the professor's accumulated professional labour?

This raises a question higher education should address before technical capability outpaces its contracts: Could a university eventually retain a reproducible version of part of a professor's teaching after it no longer employs the professor?

From analytics to ‘algorithmic management’

AI can also reshape the relationship between universities and employees by making academic work increasingly measurable. Learning management systems already capture data on faculty interactions, assessment patterns, course activity and student engagement.

Universities also maintain student evaluations, publication data, grant information and other indicators of academic performance. AI increases the capacity to aggregate, interpret and act on these data.

The International Labour Organisation uses the term ‘algorithmic management’ to describe algorithmic systems that use tracked data and other information to organise, assign, monitor, supervise and evaluate work.

Could an AI system identify allegedly underperforming instructors?

Could it compare faculty ‘productivity’?

Could it recommend workload allocations?

Could sentiment analysis of student feedback be incorporated into faculty evaluations?

Could automatically generated indicators influence contract renewal, promotion, or disciplinary proceedings?

The AAUP's report on AI and the academic professions recommends greater faculty control over educational technology, limits on intrusive monitoring and mechanisms for employees to challenge technology-related decisions.

The report’s survey included approximately 500 AAUP members from nearly 200 campuses, and 85% reported being at least somewhat concerned about the implementation of educational technology at their institutions.

British evidence predates some of the current generative AI debate and is instructive. UCU's Academic Freedom in the Digital University study surveyed more than 2,000 academics. Among its findings, 82.4% agreed or strongly agreed that digitally enabled changes to performance management had reduced academic freedom, and 83.8% said the same about digitally enabled measures of the student experience.

AI could intensify those trends.

The AFT's higher education AI principles therefore call for promotion, tenure and termination decisions to remain exclusively with human evaluators rather than with AI systems.

That is a union policy position rather than a rule that applies to every institution, but the principle is worth considering: No academic should lose employment, promotion, workload rights, or professional standing simply because an opaque system produced a score that cannot be adequately understood or challenged.

Bargaining over the AI lifecycle

Collective agreements are commonly negotiated for several years. AI systems can change significantly during the lifetime of a single agreement.

A platform purchased for one purpose may gain new capabilities through software updates. A learning management system may introduce generative functions that were not available at the time of procurement. A student support system can gradually gain analytics or automated decision-making capabilities.

Negotiating AI only at the time of procurement is therefore insufficient.

I propose treating the alternative as AI lifecycle bargaining.

This would mean establishing mechanisms for faculty and staff representatives to participate at several stages:

• Before procurement, when institutions decide whether to acquire a system.

• During implementation, when the workload, privacy and employment consequences become clearer.

• When capabilities change significantly, because software updates can transform the function of a product already in use.

• During periodic review, when the institution assesses whether a system should be expanded, modified, or discontinued.

The term ‘AI lifecycle bargaining’ is my proposed framework, not an established contractual doctrine. However, elements of this approach are already evident.

Miami's AI MOU establishes ongoing labour-management discussions. Spokane's collective agreement provides an ongoing mechanism for addressing AI issues and escalation. CUNY's 2023-27 agreement establishes a recurring labour-management committee on educational technology, with a remit that includes AI.

Universities may need to negotiate not only the adoption of technologies but also the governance of technological change over time.

The AFT's AI principles call for union involvement in identifying, procuring and administering AI systems that affect faculty and staff work. UCU likewise calls for greater transparency and accountability in AI procurement.

The logic is straightforward. By the time an expensive AI system has been purchased, integrated into university infrastructure and embedded in routine workflows, negotiation over its consequences may already be too late.

The next collective agreement

Faculty unions should not treat AI primarily as an adversary.

Used well, AI could reduce the administrative burden that has consumed academic time for decades. It could speed routine tasks, support research and give faculty more capacity for precisely the activities universities claim to value most: scholarship, mentoring, intellectual experimentation and meaningful interaction with students.

But augmentation and extraction are not the same. One model uses AI to expand the capacity of academic workers, while another uses the same productivity gains to demand more work from fewer people. Both can be described as ‘AI transformation’.

This is why the deepest labour question surrounding artificial intelligence in higher education is not whether a machine can perform what a professor once did. The more consequential questions are:

Who decides which work should be automated?

Who remains professionally accountable when AI participates in the work?

Who owns the intellectual material that AI systems depend on?

Who performs the additional labour of verification?

Who controls the data used to monitor academic work?

And who receives the productivity dividend when human labour becomes more powerful through machines?

Collective agreements cannot answer all of these questions, and bargaining rights vary considerably across countries and jurisdictions. Faculty senates, academic boards, professional associations, governments and regulators will continue to play essential roles.

But faculty unions and staff associations cannot afford to enter the conversation only after jobs have disappeared, workloads have expanded, or institutional AI systems have accumulated years of faculty-generated knowledge.

Other sectors have already faced similar issues. The Writers Guild of America’s 2026 Minimum Basic Agreement preserved the AI protections negotiated in 2023 and added requirements for companies to notify the Guild when they license writers’ work to train commercial generative-AI systems.

The Guild can request discussions that may include remuneration. SAG-AFTRA’s 2026 TV/Theatrical Agreement likewise strengthened protections for digital replicas and synthetics, building on the consent, compensation and bargaining protections negotiated in 2023.

Higher education is approaching its own version of that moment, with an important distinction.

The work now exposed to partial automation is not primarily in physical production. It includes explanation, interpretation, evaluation, advising, writing, synthesis, communication and the organisation of knowledge itself.

In other words, AI is entering the domain universities have traditionally considered academic work. The future of that work will not be determined by technological capability alone. It will also be shaped by contracts, governance arrangements, institutional choices and the bargaining power of the people whose labour is being augmented.

AI has already entered the classroom, the laboratory and the administrative office. The next place it needs to enter is the collective agreement.

James Yoonil Auh has served as vice-president of both the National Labor College in the United States and Kyung Hee Cyber University in Seoul, South Korea. He currently teaches at Judson University in Illinois. His work examines how technological and social change reshape institutions, authority and human agency.

This article is a commentary. Commentary articles are the opinions of the author and do not necessarily reflect the views of University World News.