A new academic paper is drawing attention for putting a precise economic model behind a concern that has long lived at the level of intuition: that the race to automate workforces could damage the companies running the race. The paper, "The AI Layoff Trap," by Brett Hemenway Falk of the University of Pennsylvania and Gerry Tsoukalas of Boston University, does not argue that automation is wrong in principle. It argues that the competitive dynamics of AI adoption create a structural trap from which no individual firm can escape, even when every firm can see the trap clearly.
The mechanism is not complicated, and versions of it have appeared in economic theory before. Companies replace workers with AI to reduce costs. The laid-off workers have less money to spend. Because those workers were also customers, consumer demand falls. Each company bearing a fraction of the wage bill bears only a fraction of the resulting demand loss; the cost is spread across the whole economy. So every individual decision to automate is rational, and the collective result of all those individually rational decisions is self-defeating. The paper phrases this starkly: "At the limit, this becomes self-destructive: firms automate their way to boundless productivity and zero demand."
What makes the model useful is what it says about foresight. One might imagine that firms, recognising this dynamic, would slow down. The paper finds that knowing about the trap is not enough to escape it. Competitive pressure means any firm that decelerates unilaterally falls behind competitors who do not. The result is what the authors call an "automation arms race," driven not by ignorance of the consequences but by the structure of competition itself. The incentives are asymmetric: the benefit of automation is captured privately, while the cost in reduced consumer spending is distributed across the economy.
The numbers the paper points to are not theoretical. More than 100,000 technology workers were laid off in 2025 with AI cited as a key driver in over half the cases. In the first months of 2026, that number is already above 90,000 across nearly 100 companies. The paper evaluates the standard policy responses, including universal basic income, reskilling programmes, and worker equity participation, and finds them insufficient. Its conclusion is direct: only a targeted tax on automation, a Pigouvian tax designed to make companies internalise the demand destruction their layoffs create, can correct the distortion. The paper does not hold that such a tax is politically likely; it holds that it is the only mechanism that addresses the actual problem.
The framing sits in interesting tension with a report released the same week by Singapore's Ministry of Manpower. Surveying 2,560 firms employing close to 500,000 workers, the ministry found no indication of widespread job displacement attributable to AI so far. Headcount reductions were reported by 6.2 percent of firms actively using AI, while 70.7 percent of those firms reported productivity gains. Singapore's MOM concluded that "early evidence suggests AI is complementing rather than displacing labour."
The two accounts are not directly contradictory. Singapore's data captures the current moment, with AI adoption still limited to fewer than 30 percent of firms. The "AI Layoff Trap" paper models what happens as that adoption spreads and intensifies under competitive pressure, particularly as AI systems become more capable. The Singapore report itself notes that "workers who do not upskill risk being left behind as AI transforms our day-to-day tasks," and that smaller firms already face structural disadvantage in keeping pace. The authors of the trap paper would likely observe that the complement-versus-displacement question is a function of intensity and time, not a stable equilibrium.
What the paper contributes is a formal account of why market forces alone will not produce a good outcome even when all the participants understand the problem. The mechanism it describes is one where competition functions not as a correction but as a trap. Whether policymakers engage seriously with that argument is a separate question. The current political trajectory in most economies suggests they will not, at least not yet.