Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
AI models can diagnose cancer, draft contracts and beat the world’s best competitive coders. But can they generate tax revenue? The question is increasingly pressing for governments around the world as anxiety about job automation grows.
At present, personal income tax, which is levied primarily on employees’ wages, accounts for almost one-quarter of the total tax take across OECD countries. Social security contributions, paid by both employers and employees, account for a similar amount. The US, which has no national consumption tax, relies even more heavily on labour taxation.
So if AI does unleash job market destruction, governments could face a sudden shortfall in the revenues they need to help pick up the pieces. Leaders could also face greater calls to support workers through the transition, whether via existing welfare and job support systems, intensive reskilling schemes, or even by cutting wage-related taxes to tilt the scales in employees’ favour.
Income tax could be obsolete within five years, Tom Blomfield, founder of the digital bank Monzo, asserted last month. “I don’t think we’ll tax human labour, we’ll tax compute,” he said.
That is the worry fuelling fresh calls for a rethink of the traditional model of taxation.
OpenAI published a think piece last month that it framed as the “starting point for a broader conversation about how to ensure that AI benefits everyone”. It suggested that states should consider higher taxes on capital gains and corporate profits, levies targeting AI-driven returns and incentives for companies to retain and retrain workers.
These ideas, increasingly popular with AI evangelists, are gaining traction with US Democrats too. Senator Bernie Sanders last year called for a “robot tax” on each job replaced by automation. Others favour so-called “token taxes” on the use of AI models or direct state investment in AI companies.
Of course, a radical rethink may prove unnecessary. In the most utopian scenario, AI could put governments in a better fiscal position, noted Arun Advani, director of Warwick university’s Centre for the Analysis of Taxation. If the technology makes workers more productive rather than replacing them, it could raise both wages and consumption, helping governments run public services at a lower cost.
Most economists now think it probable that at least some workers will lose out — at least in the short term. But the gains generated by AI could still be large enough for governments to repair the damage using existing tools.
New modelling by researchers at the Windfall Trust, intended to illustrate how AI-related job losses might erode tax bases, suggests the net impact for an “average” OECD economy might be near-neutral in a “low impact” scenario. It assumes 12 per cent of workers are displaced, such as translators, junior developers and paralegals, but that AI productivity gains are largely retained by domestic companies and consumers.
“It was a little less disruptive than we might have expected,” said Deric Cheng, one of the paper’s authors.
Governments facing this challenge could reach for off-the-shelf solutions. If AI made many goods and services cheaper, “you could do a lot with a consumption tax”, Warwick’s Advani argued. The usual objection — that consumption taxes hit those on lower incomes — could be less of a worry if AI hit higher-paid white-collar workers and eroded wage differentials. Moreover, sales taxes are hard to avoid.
Policymakers could also move to equalise labour income and capital gains tax, as inequality campaigners have long wanted. This is harder: capital gains are generally taxed more lightly than wages at present, because of wealthy individuals’ ability to move overseas and worries about deterring saving. But land and inheritance taxes could also come into play.
The real difficulties will arise if AI destroys jobs and generates large monopoly profits that flow beyond the taxman’s reach offshore, most likely to US tech giants. President Donald Trump’s rejection of an OECD-brokered deal to set a global minimum tax on corporate profits shows how difficult it would then be to find an international response.
So this is where more ingenious ideas for taxes directly targeting the deployment or use of AI might come into play.
Previous calls for “robot taxes”, mooted by Bill Gates as long ago as 2017, met resistance. This is because there is no clear way to define a robot. Nor is there reason to penalise one labour-saving device more than another, especially if it is intended to raise productivity and generate wealth.
So-called “token taxes” — applied to the tokens generated by AI models at the point of sale, already used as a basis for billing by AI providers — are another idea, floated by Anthropic chief Dario Amodei among others. Other suggestions include new versions of “robot taxes” — imposing steeper tax rates on companies that deploy AI or reap very high profits from it.
These could create distortions, penalise particular forms of innovation, impede productivity gains and potentially push companies to relocate to friendlier tax jurisdictions. But they should not be ruled out.
The hope is that workers will be able to harness AI, and that its gains will be widely shared — with enough competition to prevent them from being captured by an elite few. But as Advani noted, there is enough risk of a “crazy giant shock” that governments need to prepare — and fast.
Other readables
Free Lunch is edited by Harvey Nriapia