To understand why AI will change the labor market, the policy world must first consider why workers have historically held leverage. Workers have primarily derived their value to employers from the products and services they produce, their support for company operations, and the social skills and relationships they bring to the workplace. Additional sources of leverage have included partial company ownership, contractual protections, and specialized knowledge of a company. Together, workers have also exercised collective power through their ability to organize and withhold their labor.
While employers have always been able to threaten to fire a worker for poor performance, maintaining the same level of production has often required hiring someone else to fill that gap. The introduction of AI into company workflows threatens this logic because humans are no longer competing only against one another. They are also competing against AI software and robots. In some cases, AI can be cheaper, more efficient, and capable of producing higher-quality work. While the effects will vary across fields, competitive pressure may push companies to integrate AI even when doing so reduces their need for human workers.
This change threatens both the leverage workers receive from being skilled at their jobs and their collective power, as some companies may become less dependent on them. With the possibility of this massive shift against labor comes a responsibility for the government to protect workers’ bargaining power and make human employment more attractive. While AI may be efficient, replacing a human should not become a no-brainer because the tax code makes employing people disproportionately expensive. The decision should depend more on the quality and productivity of the work and less on unequal tax treatment.
One way to address this problem is to redesign the tax code so that it relies less on labor and more on other sources of economic value, including services and capital income. If AI reduces the share of income going to workers, a government heavily dependent on taxing that income could face growing fiscal pressure. However, this transition must be handled carefully. The government should protect workers while preserving incentives to innovate and invest.
The first step should be to ensure that human labor is not disproportionately taxed compared with automated alternatives. Under the Federal Insurance Contributions Act, employees and employers generally each pay 7.65 percent in payroll taxes: 6.2 percent for Social Security and 1.45 percent for Medicare. The Social Security portion applies only up to an annual earnings limit, while Medicare has no equivalent cap. Some higher-income workers also pay an additional Medicare tax that employers do not match. These qualifications mean that 15.3 percent is the standard combined rate below the Social Security earnings limit, rather than a flat rate on all wages. (Internal Revenue Service)
The difference extends beyond payroll taxes. In a 2020 study, economists Daron Acemoglu, Andrea Manera, and Pascual Restrepo estimated an effective tax rate on labor of approximately 25.5 percent, compared with roughly 5 percent on equipment and software following the 2017 tax reform. These are historical estimates of effective tax burdens, rather than current statutory rates, but they demonstrate how the tax system can favor investment in automation over employing people. (Acemoglu et al. 259–60)
Meanwhile, a September 2026 RAND commentary, citing a recent study, reported that nearly two-thirds of federal revenue comes from taxes on human labor. The government therefore depends heavily on the same workers whose employment can be disadvantaged by parts of the tax code. This does not mean that automation is entirely untaxed; business profits and investment income can still face taxes. It means that employing a person carries specific tax obligations that purchasing an automated alternative does not necessarily carry. (“Preserving Jobs”)
The response should combine lower payroll taxes with new revenue from economic activity increasingly performed through automation. However, the amount of money involved is substantial. USAFacts reports that payroll taxes accounted for 33.6 percent of the federal government’s $5.26 trillion in revenue in fiscal year 2025, or approximately $1.77 trillion. Replacing all payroll-tax revenue would therefore require an enormous alternative funding source. (USAFacts)
A more focused starting point would be to reduce the employer share of Social Security and Medicare taxes while retaining the standard employee contribution. This would directly reduce the added cost of hiring a worker. It would also require less replacement revenue than eliminating both sides, although the exact cost would need to be estimated rather than assumed to equal half of all payroll-tax collections.
So, where could that replacement revenue come from? In a Brookings article, Anton Korinek and Lee M. Lockwood discuss five approaches:
- Digital services taxes: taxes on services provided through AI.
- Token taxes: charges based on units of AI-generated output.
- Robot service taxes: taxes on services delivered by robots.
- Robot taxes: taxes on owning or operating robotic equipment.
- Compute taxes: taxes on the computational resources used by AI.
Their central distinction is between taxing services purchased by final consumers and taxing the productive assets used to provide them. (Korinek and Lockwood)
A combination of revenue sources would help avoid placing too much of the burden on one area. Further research is necessary to determine appropriate rates, but I believe the most useful starting points are a digital services tax and a robot service tax. These would collect revenue from the services automation produces rather than directly taxing the purchase of the technology itself.
There is already precedent for taxing digital businesses, although existing laws differ from this proposal. The United Kingdom, for example, introduced a 2 percent tax on specified revenues from search engines, social media services, and online marketplaces. That is not a general tax on AI services, but it provides an example of a government defining and taxing particular forms of digital economic activity. (HM Revenue & Customs)
Under my proposal, both service taxes should apply only to companies bringing in more than $100 million in annual profit. This would allow smaller and less profitable companies to continue developing and using new technology without facing the same burden as companies that have already reached substantial profitability. To prevent avoidance through the creation of smaller subsidiaries, the threshold should be measured across the parent company and its controlled businesses. The tax should also phase in as profits rise above $100 million, rather than suddenly imposing the full burden when a company crosses that line.
However, the profit threshold and the thing being taxed are two different parts of this framework. Profit would determine whether a company qualifies to pay, while revenue from covered services sold to U.S. consumers would determine how much it owes. The law would need a consistent definition of profit, rules against shifting it between related companies, and clear reporting requirements.
For the digital services tax, covered services could include paid AI assistants, automated content generation, and other digital services in which AI performs the main task the customer is purchasing. A company should not have its entire revenue taxed simply because its employees use AI. An architect using AI to help design a building is different from a company selling an entirely automated design service. The government would need to establish clear categories based on what customers are buying, with rules for separating the automated service from other parts of a bundled product.
The robot service tax would apply a similar principle to physical work. My proposed categories would include autonomous transportation, robotic delivery, and automated cleaning sold to final customers. The taxable amount would be the payment for the service. A company buying a robot would not owe this tax merely because it purchased the equipment. If a service uses both AI software and a physical robot, the same sale should be taxed only once.
Both taxes should generally apply at the final sale, with exemptions for verified business purchases used to produce another good or service. Otherwise, the same economic activity could be taxed repeatedly as it moves between companies, raising costs beyond the intended rate. This follows the Brookings framework’s emphasis on distinguishing final consumption from business investment. (Korinek and Lockwood)
This approach does have a limitation: it would not directly tax every instance of a company replacing a worker. A business using its own AI for accounting, for example, would not necessarily make a separate taxable sale. These taxes would help shift the government’s revenue base toward automated services, but they would not establish equal tax treatment in every hiring decision. Reducing employer payroll taxes would address the hiring side of the proposal more directly.
The rates should initially be modest and introduced over several years. As an illustration, a 3 percent tax on $1 billion in covered sales would raise $30 million before exemptions and changes in customer behavior. This is not a revenue forecast or a claim that 3 percent is the correct rate. It shows why the size of the taxable market matters as much as the percentage chosen. Even a 5 percent tax would require $20 trillion in taxable sales to generate $1 trillion before accounting for exemptions or changes in behavior. The government cannot responsibly assume that two narrowly defined taxes, especially with a $100 million profit threshold, will immediately replace the payroll tax.
For this reason, the first goal should be a gradual reduction in the employer portion of payroll taxes, with each reduction backed by dependable replacement revenue. Congress should require independent estimates before setting the rates and review actual collections before authorizing further reductions. If the two service taxes cannot support full elimination, the government would need either a smaller payroll reduction or additional revenue from a broader base, including corporate profits and capital income.
The legislation must also protect Social Security and Medicare. Revenue replacing their payroll contributions should be directed to the relevant trust funds, and workers should continue receiving credit for their covered earnings. Decreasing the cost of hiring should not come at the expense of the benefits those workers have earned.
Finally, charging companies a tax does not guarantee that companies bear its entire cost. Some of it could be passed on through higher prices. A portion of the revenue should therefore support a refundable credit for lower-income households, with that cost accounted for before calculating how much payroll relief the proposal can fund. Foreign and domestic providers selling covered services to U.S. customers should face the same rules, so moving a company’s headquarters abroad does not remove the obligation.
Ultimately, the tax code should evolve alongside the economy it is meant to support. If companies become less dependent on human labor, the government cannot continue depending on that labor to the same degree for its revenue. A gradual shift toward taxing the value sold through automated services would help address this problem while making human employment less expensive. AI can increase what the economy produces, but public policy must ensure that workers also benefit from that increase.
Works Cited
Acemoglu, Daron, et al. “Does the US Tax Code Favor Automation?” Brookings Papers on Economic Activity, Spring 2020, pp. 231–300. https://www.brookings.edu/wp-content/uploads/2020/12/Acemoglu-FINAL-WEB.pdf.
HM Revenue & Customs. “Digital Services Tax.” GOV.UK, 11 Mar. 2020, https://www.gov.uk/government/publications/introduction-of-the-digital-services-tax/digital-services-tax.
Internal Revenue Service. “Topic No. 751, Social Security and Medicare Withholding Rates.” IRS, https://www.irs.gov/taxtopics/tc751. Accessed 26 Oct. 2026.
Korinek, Anton, and Lee M. Lockwood. “The Future of Tax Policy: A Public Finance Framework for the Age of AI.” Brookings, 8 Jan. 2026, https://www.brookings.edu/articles/future-tax-policy-a-public-finance-framework-for-the-age-of-ai/.
“Preserving Jobs and Tax Revenue in the Age of AI: Q&A with Carter Price.” RAND, 17 Sept. 2026, https://www.rand.org/pubs/commentary/2026/09/preserving-jobs-and-tax-revenue-in-the-age-of-ai-qa.html.
USAFacts. “How Much Money Does the US Federal Government Collect?” USAFacts, 16 Oct. 2026, https://usafacts.org/answers/how-much-does-the-us-federal-government-collect/country/united-states/. Accessed 26 Oct. 2026.