Consider this scenario.
A small business owner in London replaces their human personal assistant, accountant, web designer and social media manager with AI agents. Four different roles, but zero salaries (PAYE), zero National Insurance, and zero pension contributions.
This transition is occurring today, at scale. The deeper question is whether our tax frameworks, which were built for economies anchored in human labour, can adapt before the new AI "workforce" begins impacting the fiscal foundations of welfare states.
The Scale of What Is Already at Stake
In 2024 to 2025 in the UK, Income Tax, Capital Gains Tax and National Insurance contributions together represented 57% of all HMRC annual receipts, totalling £475.3 billion. These revenues fund the NHS, social care, defence, schools, and the entire infrastructure of the UK economy.
The modern tax system rests primarily on two pillars: labour income and, to a lesser extent, consumption. In the United States, approximately three quarters of all federal tax revenue derives from labour. The UK's architecture is structurally identical. Generative AI now threatens the first pillar by reducing demand for human work across precisely the knowledge-worker occupations that generate the bulk of income tax receipts.
The exposure data is stark. Goldman Sachs Research (2023) estimates 300 million full-time jobs globally are exposed to automation. In the United States and Europe, approximately two-thirds of current jobs face some degree of exposure, while up to one quarter of all work tasks could be performed entirely by AI. Office and administrative support roles show the highest exposure at 46%, followed closely by legal work at 44%. McKinsey Global Institute (2017) projected that automation could require up to 375 million workers worldwide to switch occupational categories by 2030. In another estimate, the World Economic Forum Future of Jobs Report (2025) highlighted that AI will displace 92 million jobs by 2030.
While AI promises substantial productivity gains that could expand other tax bases such as corporate profits, VAT and other revenue streams, the transitional pressure on public finances, especially concurrent demand for social safety nets, remains acute.
The Fundamental Inequity
A human accountant earning £55,000 per year pays income tax and National Insurance. Their employer pays employer NICs on top. The state collects, conservatively, upwards of £20,000 per year in combined fiscal obligations from that one employment relationship. In 2025–26, National Insurance Contributions alone are estimated to raise £205.4 billion — representing 16.7% of all HMRC receipts and equivalent to around £7,150 per household.
Now replace that accountant with an AI agent subscription. The cost to the business might be £100-200 per month. The tax collected by HMRC? Zero. The NI paid? Zero. The pension contribution? Zero. The work is identical but the economic displacement is real. The fiscal contribution is non-existent.
Human Accountant (£55k/yr) vs AI Agent (same output):
✅ Income Tax — pays up to 45% | ❌ AI Agent — zero
✅ National Insurance — pays employee + employer | ❌ AI Agent — zero
✅ Pension Contribution — auto-enrolled, employer matched | ❌ AI Agent — zero
✅ Employment Rights — holidays, sick pay, maternity | ❌ AI Agent — none
✅ HMRC PAYE Visibility — full audit trail | ❌ AI Agent — invisible
Multiplied across millions of roles, this structural subsidy for AI labour makes it categorically cheaper, not because it is inherently superior in every context, but because it evades the fiscal obligations applied to humans.
Bill Gates articulated this logic directly in 2017: "Right now, the human worker who does $50,000 worth of work in a factory — that income is taxed. If a robot comes in to do the same thing, you'd think that we'd tax the robot at a similar level."
Eight years on, Gates's observation has compounded in urgency. AI agents are not industrial robots confined to factory floors. They operate in every office, every profession, every sector. They are invisible, scalable, and structurally untaxed. We are quietly constructing the world's first tax-free labour force — invisible, scalable, and growing every quarter.
Why Previous Approaches Failed and What We Must Learn
In February 2017, the European Parliament considered a proposal to tax robot owners to fund retraining for displaced workers. The proposal was rejected — hailed by the robotics industry as avoiding an innovation penalty. EU Commissioner Andrus Ansip stated that any jurisdiction implementing one would become less competitive as technology companies were incentivised to move elsewhere.
South Korea, however, took a different route. In August 2017, under President Moon, it passed what became known as the world's first robot tax — reducing tax breaks previously awarded to investments into robotics. Modest. Indirect. But it established a precedent: fiscal systems can and do respond to automation. In the UK in 2017, some legislators called for a robot tax, but it went nowhere.
The three recurring objections were: definitional ambiguity, jurisdictional arbitrage, and risk to innovation.
These concerns carry far less weight for today's AI agents, which are software — jurisdiction-agnostic and already deeply embedded in commercial life. Distinguishing substitution (AI performing an entire role) from augmentation (tools enhancing human productivity) is feasible under existing employment classification rules. Taxing substitution, not innovation itself, avoids penalising technological progress.
Four Policy Frameworks Worth Examining
No single mechanism is without flaw. All four merit serious parliamentary and regulatory scrutiny.
1. Employer Substitution Surcharge — A National Insurance–equivalent levy triggered when an AI agent directly substitutes a human role. Calculated against median salaries for standardised occupational categories. New legislation will be required.
2. Compute Tax — A modest levy on commercial AI processing power consumed above a defined threshold, applied at data-centre or cloud-provider level. Technology-neutral and difficult to avoid.
3. Output VAT Extension — Extend VAT obligations to AI-generated commercial outputs (legal summaries, financial reports, marketing content). Aligns with Brookings Institution (January 2026) analysis that consumption taxation must carry greater weight as labour taxes erode.
4. Windfall Profits Levy — Higher corporation tax rate on profits demonstrably attributable to AI-driven headcount reductions, with mandatory reporting.
None of these policy mechanisms are ready for immediate implementation without further policy development. All are more sophisticated than the blunt robot taxes that failed in 2017. The question is whether government has the institutional appetite to develop them before the fiscal impact becomes irreversible. A pragmatic hybrid — mandatory HMRC disclosure of AI-related headcount changes, plus a phased substitution surcharge combined with VAT extension — offers the clearest path forward.
The Case Against Taxing AI Agents
Not everyone agrees that taxing AI agents is the right response, and the counterarguments deserve serious engagement.
From a capital markets perspective, any fiscal levy on AI deployment risks repricing the cost of automation at precisely the moment the UK needs to attract technology investment. Institutional investors and venture capital allocators are acutely sensitive to jurisdiction-specific regulatory costs; an AI surcharge that does not exist in the United States, Singapore, or the UAE creates an immediate arbitrage incentive to domicile AI-intensive operations elsewhere. EU Commissioner Andrus Ansip made exactly this point when rejecting the 2017 robot tax proposal, and the structural logic has not changed. The International Federation of Robotics observed that countries with the highest automation rates also tend to have the lowest unemployment rates, suggesting that taxing productive capital may suppress the very output growth that expands other tax bases over time.
On workforce planning and innovation, the argument against is equally pointed. AI agents do not simply substitute for human labour; rather, in many deployments they augment it, enabling smaller teams to operate at the scale of larger ones, allowing SMEs to compete with corporates, and freeing skilled workers from administrative burden to focus on higher-value activity. A blunt substitution tax risks penalising augmentation alongside replacement, embedding a definitional ambiguity that compliance teams will exploit and tribunals will spend years resolving. The Congressional Budget Office (2024) noted that AI-driven productivity growth could raise federal revenues and reduce deficits, suggesting the fiscal system may self-correct through expanded corporate profits and VAT receipts without requiring a dedicated AI levy at all. The strongest version of this argument is not that AI should never be taxed, but that poorly designed taxation now could lock in competitive disadvantage and slow the productivity gains, which for many countries represent the best medium-term fiscal option.
The Ethical and Moral Dimension
There is also a deeper moral tension at the heart of this debate that purely fiscal arguments tend to obscure. Taxing AI agents implies, even implicitly, that they occupy a category analogous to workers; entities whose economic contribution carries social obligation. But AI agents have no lived experience, no dependents, no retirement, no illness, and no stake in the society whose infrastructure they are deemed to be underfunding. To construct a tax framework around them risks anthropomorphising tools, creating legal and philosophical precedents with consequences far beyond fiscal policy. The more uncomfortable moral question is not whether AI should pay tax, but why human labour is taxed so heavily in the first place, and whether the current system, which loads the cost of the welfare state disproportionately onto earned income, was ever the right design. If AI forces a reckoning with that structural inequity, the ethical response may be to redesign the tax base entirely rather than extend a flawed model to machines.
There is also a justice argument that cuts the other way from the one usually made. Access to AI agents is not the exclusive preserve of large corporations; it is rapidly becoming available to sole traders, micro-businesses, disabled entrepreneurs, and individuals in the developing world who have never had access to professional services at all. A tax on AI deployment, however well-intentioned, risks pricing out precisely those who stand to benefit most from democratised access to legal, financial, and administrative capability. If the moral imperative is to protect the vulnerable, taxing the tool that is reducing inequality of access to professional expertise requires a more careful ethical justification than the fiscal argument alone provides.
The Deeper Question: Who Bears the Cost of the Transition?
The fiscal argument for AI taxation is not, at its core, about punishing innovation. It is about distributional justice and fiscal sustainability. When AI displaces a worker, the productivity dividend accrues to the firm and its investors. Retraining, welfare and housing costs fall on the public purse. This is not a market failure but rather a regulatory and fiscal gap: a hidden subsidy making AI labour categorically cheaper. Early intervention, before autonomous systems fully embed in commercial infrastructure, is far more tractable than retrospective taxation of an already-transformed economy. Timing is not a secondary consideration. It is the primary one.
What Countries and Economies Should Do Now
No single jurisdiction can solve this alone, and the history of robot tax proposals demonstrates precisely what happens when one country moves unilaterally while others wait. The most urgent priority for any advanced economy is the same: measure before you legislate. Mandatory disclosure of AI-related headcount changes in corporate tax filings costs nothing to implement, requires no new legal framework, and immediately creates the evidence base without which any policy intervention is guesswork. The United States, the United Kingdom, the European Union, and OECD member states should move in concert to standardise this disclosure requirement, just as they have done with country-by-country tax reporting under BEPS. Without coordinated measurement, governments will be debating estimates while the fiscal gap widens.
For advanced economies with mature tax administrations such as the UK, Germany, France, the United States, Japan, Canada, and Australia, the immediate priority beyond disclosure is scoped fiscal modelling: a quantified assessment of the labour tax exposure created by AI substitution in professional services, conducted with OECD coordination and a twelve-month mandate. These economies carry the highest concentration of knowledge-worker roles most exposed to AI displacement, and they have the institutional capacity to act. The Brookings Institution's January 2026 analysis is clear that the window for tractable intervention is narrowing, and that consumption-based frameworks (including an Output VAT Extension) represent the most internationally portable mechanism given existing VAT and GST infrastructure across these jurisdictions.
For emerging and middle-income economies where AI adoption is accelerating but formal employment and tax collection infrastructure is already fragile, the calculus is different. The World Economic Forum's Future of Jobs Report (2025) projects 92 million job displacements by 2030, with developing markets facing acute exposure in administrative and services sectors without the social safety net capacity of advanced economies. For these countries, the priority is not AI taxation but AI transition financing — multilateral mechanisms, potentially through the IMF, World Bank, or a dedicated OECD facility, that fund retraining and social protection during the displacement period. Taxing AI prematurely in contexts where it is also a primary driver of economic inclusion would be self-defeating.
At the international level, the G20, which successfully coordinated the global minimum corporate tax framework under Pillar Two, is the natural forum for an AI fiscal compact. A compact with three components would be achievable within a two-year horizon: standardised AI headcount disclosure, a coordinated framework for distinguishing taxable substitution from augmentation, and a burden-sharing mechanism for transition financing in lower-income economies. The technology is not waiting for policy consensus. The question is whether the institutions that govern the global economy have the coordination capacity and political will to respond before the fiscal damage becomes structural and irreversible.
Conclusion: A New Economy Is Emerging — The Question Is Who Designs It
We are not simply witnessing a disruption to the labour market. We are witnessing the emergence of an entirely new economic order, one in which the traditional compact between work, contribution, and social protection is being rewritten in real time, without democratic deliberation, and without fiscal design.
Every previous technological revolution — steam, electricity, computing — ultimately expanded the tax base, created new categories of work, and preserved the social contract, even through painful transitions. AI may do the same. But those earlier revolutions did not arrive with the speed, the scale, or the invisibility of software deployed globally overnight. And they did not hollow out the knowledge-worker class, the very taxpayers on whom modern welfare states most depend - simultaneously and across every sector.
The societies that thrive in this new economy will not be those that taxed AI most aggressively, nor those that refused to act at all. They will be those that had the foresight to redesign their fiscal architecture before the old one collapsed, shifting from taxing human effort to taxing economic value, wherever and however it is generated. That is not a radical proposition. It is the logical conclusion of applying the principles that have always governed fair taxation to the economic reality that is already here.
The tax-free AI workforce is not a loophole to be closed. It is a signal that the entire framework needs reimagining. The question for governments, parliaments, and international institutions is no longer whether to act - it is whether they can move with sufficient clarity and coordination to shape the new AI economy, rather than inherit it.
_______________________________
Prof. Dr. Naseem Naqvi MBE FBBA
President, The British Blockchain Association
Prof. Dr. Naseem Naqvi MBE FBBA is President of the British Blockchain Association, Editor-in-Chief of the Journal of the British Blockchain Association, and Head of Secretariat for the All-Party Parliamentary Group on Blockchain Technologies. He coined the term Evidence-Based Blockchain (EBB) in 2018 and holds the distinction of being the first person globally to receive a UK National Honour for services to blockchain and digital asset technologies.
Data Sources: HMRC Annual Bulletin 2024/25 · OBR Economic and Fiscal Outlook, November 2025 · Goldman Sachs Research (2023) · World Economic Forum Future of Jobs Report (2025) · Brookings Institution (January 2026) · McKinsey Global Institute · House of Commons Library Tax Statistics, March 2026.