Across much of the world, tax systems were built on a simple economic assumption: most people earn their living through work, and most public revenue comes from taxing that work. Wages are visible, payroll is trackable, and employment is the engine that turns households’ income into government budgets. But as artificial intelligence accelerates automation and reshapes how tasks are performed, governments are beginning to confront a less comfortable possibility: if AI reduces job markets or changes the nature of employment faster than tax policy can adapt, labour-income tax receipts could fall—sometimes sharply—and public services would face a funding gap.
The Financial Times discussion captured the core tension: if AI destroys or hollowes out labour markets, the state may lose one of its most reliable revenue streams. That doesn’t automatically mean governments will run out of money. It means they will have to replace labour taxes with something else, and do so while maintaining legitimacy, fairness, and economic stability. The challenge is not only fiscal; it is political and administrative. Tax bases are not just numbers in a spreadsheet—they are relationships between citizens, firms, and the state. When those relationships shift, the entire architecture of public finance must be reconsidered.
What makes the problem urgent is that the timing may be misaligned. Labour-market disruption can happen quickly—through hiring freezes, task substitution, and productivity gains that reduce the need for human labour. Tax policy, by contrast, often moves slowly. Legislative cycles, court challenges, and the complexity of designing new taxes or reforming existing ones can take years. In the interim, governments may experience a “revenue shock”: a decline in wage-related receipts before replacement mechanisms are ready.
To understand why labour-income taxes are vulnerable, it helps to look at what they actually measure. Payroll taxes and income taxes tied to wages depend on three variables: the number of workers, the level of wages, and the share of total income that takes the form of taxable employment earnings rather than capital income, benefits, or untaxed transfers. AI can affect all three. Even if overall economic output rises, the distribution of income can change. If more value accrues to owners of capital, to firms that deploy AI, or to highly skilled workers while routine roles shrink, labour’s share of income can fall. That means fewer taxable wages, even in a growing economy.
There is also a structural issue: many tax systems are designed around stable employment relationships. When work becomes more fragmented—through gig arrangements, short-term contracts, platform-mediated labour, or “task-based” contracting—compliance and collection can become harder. Some income may fall outside traditional payroll channels. Others may be taxed less effectively due to classification disputes or enforcement gaps. AI can intensify these trends by making it easier for firms to outsource tasks, automate parts of service delivery, and rely on flexible labour models. The result is not only fewer jobs, but different jobs—jobs that may be harder to tax in the same way.
So what happens when labour-income tax receipts shrink? The immediate answer is straightforward: governments must find alternative revenue sources or cut spending. But the real question is which alternatives are feasible without undermining growth, social cohesion, or investment incentives. The debate is already moving toward a set of options that, taken together, amount to a broader rethinking of how states fund themselves in an automated world.
One path is to broaden existing tax bases. If labour taxes decline, governments can attempt to shift toward consumption taxes such as VAT or sales taxes. Consumption taxes are often easier to collect because they attach to transactions across the economy, not to employment status. In theory, if AI increases productivity and lowers costs, consumption may rise, sustaining revenue. In practice, however, consumption taxes can be regressive unless paired with targeted rebates or social transfers. In countries where inequality is already politically sensitive, raising VAT without compensation can trigger backlash. Governments would need to design offsetting measures—perhaps expanding cash transfers, lowering taxes for low-income households, or using automatic stabilisers to protect vulnerable groups.
Another approach is to increase taxes on capital or profits. If AI concentrates economic gains among firms and asset owners, then corporate taxation and capital taxation could become more important. Yet this route is complicated by global competition and the mobility of capital. Firms can shift profits across jurisdictions, and tax authorities face the challenge of attributing income to specific locations when AI systems and data-driven business models operate across borders. Even within a single country, the question of what counts as “profit” becomes more complex when intangible assets—algorithms, models, datasets, and intellectual property—drive value creation. Traditional accounting rules may not capture the economic reality of AI-driven businesses, leading to either under-taxation or disputes.
That is why some proposals focus on “taxing the value created by automation” rather than simply taxing profits. The idea is to identify a measurable base linked to AI deployment—such as the use of automated systems, the reduction in labour costs, or the share of output produced with minimal human input. But translating that into enforceable policy is difficult. Companies can argue that automation is not a discrete input but a continuum of process improvement. They can also claim that AI complements labour rather than replaces it, or that labour displacement is temporary. Enforcement would require clear definitions, robust reporting standards, and audit capacity that many tax administrations may not yet have.
A related concept is a levy on “robotics” or automation. These proposals have appeared in various forms over the years, often with the aim of capturing revenue from productivity-enhancing technologies that displace workers. The problem is that automation is not a single technology category. It includes software, industrial equipment, logistics optimisation, algorithmic decision-making, and increasingly AI-enabled systems embedded throughout supply chains. A narrow robot tax risks being both arbitrary and easy to circumvent. A broad automation tax risks being economically disruptive and administratively unmanageable. Policymakers would need to decide whether the tax targets the technology itself, the economic effect, or the distributional outcome—each choice carries trade-offs.
Some governments may instead pursue a “social insurance” model that decouples benefits from employment. If labour markets shrink, unemployment insurance and welfare systems still need funding. One way to address the revenue gap is to create new contributions based on economic activity rather than payroll. For example, levies could be tied to company revenues, value added, or even the use of AI-enabled services. This resembles a shift from taxing labour to taxing the economic capacity that replaces labour. The advantage is that it aligns funding with the source of economic value. The disadvantage is that it can blur the line between tax and regulation, and it may raise costs for firms that are trying to invest and compete.
There is also the possibility of using sovereign wealth-style approaches. If AI-driven productivity increases national income, governments could capture a portion through public investment funds, similar to how some countries manage oil revenues. But unlike natural resources, AI value is not a finite commodity. It is generated continuously by private firms and depends on innovation ecosystems. Public funds would need a credible mechanism to acquire returns—through direct ownership, profit-sharing arrangements, or revenue earmarking. That is politically attractive in theory but challenging in practice, especially when budgets are already strained.
Perhaps the most politically sensitive option is to adjust labour taxes themselves. If employment declines, raising labour tax rates might seem like a way to maintain revenue per worker. But that can worsen the problem by increasing the cost of hiring and encouraging further automation. It can also deepen inequality if high-skilled workers remain employed while lower-skilled roles disappear. Governments could instead reduce labour taxes to encourage employment, but that would likely widen the revenue gap unless offset elsewhere. In other words, labour tax reform is not a standalone solution; it is part of a broader package.
A unique angle emerging in policy discussions is the idea of treating AI as a factor of production that should contribute to the social contract. In traditional economics, land, labour, and capital each have a role. If AI becomes a pervasive production factor—like capital—then it may be reasonable to ask whether it should be taxed similarly. But AI does not fit neatly into existing categories. It is partly software, partly data-driven capability, partly organisational know-how, and partly a service delivered through platforms. Tax systems struggle when the object of taxation is intangible and distributed.
This is where administrative capacity becomes central. Even the best-designed tax policy can fail if it cannot be implemented. Governments may need new reporting requirements for AI deployment, improved data sharing between tax authorities and regulators, and better tools for detecting under-reporting. They may also need to modernise tax administration to handle real-time economic activity. The irony is that AI could help governments collect taxes more effectively—using analytics to detect anomalies, automate compliance checks, and reduce fraud. But that requires investment and careful governance to avoid privacy violations and bias.
Another dimension is the labour market transition itself. If AI displaces workers, the state’s fiscal challenge is not only revenue loss; it is also increased spending needs. Unemployment benefits, retraining programmes, disability support, and social services may rise even as labour tax receipts fall. That means the budget gap could be larger than the revenue decline alone suggests. Governments may face a double hit: less income from taxes and more demand for public support.
This is why some analysts argue that the policy response should not be framed solely as “finding new taxes.” It should be framed as building a new social and economic bargain. If AI reduces the number of people who earn wages, then societies may need to ensure that income support is not tied exclusively to employment. That could involve expanding universal basic income-like schemes, negative income taxes, or wage insurance models. Funding those programmes would still require revenue, but the design could reduce the political pressure to punish automation through blunt taxes. Instead, the state could aim to share productivity gains broadly while maintaining incentives for innovation.
The question then becomes: where do productivity gains go? In many economies, AI adoption could increase profits and concentrate wealth. If labour’s bargaining power weakens, wage growth may lag behind productivity. In that scenario, labour-income taxes fall not only because fewer people