The debate over artificial intelligence is often cast in extremes. On one side are the enthusiasts, who speak as if every technological advance justifies itself and every displaced worker is simply part of the cost of progress. On the other are the alarmists, who talk as though the labor market is on the verge of overnight collapse. Both reactions miss the deeper issue. The real question is not whether AI will change work. That is already happening. The real question is whether the countries being changed by it have built any serious protections for the people expected to absorb the shock. Right now, the evidence suggests the world is not prepared. The World Economic Forum said in January that job disruption will affect 22% of today’s jobs by 2030, with 170 million new roles created and 92 million displaced, while 40% of employers say they expect to reduce their workforce where AI can automate tasks. That is not apocalypse. But it is serious enough to call for more than optimism.
The job losses are no longer theoretical, even if the causes are sometimes messier than press releases suggest. Reuters reported in March that investors’ and economists’ concerns are deepening and that job losses are already emerging in sectors most exposed to automation. Goldman Sachs economists estimated that AI contributed to between 5,000 and 10,000 net job losses each month last year in the most exposed U.S. industries, while Challenger, Gray & Christmas said AI accounted for 7% of the planned layoffs announced in the United States in January. Companies from Amazon to Pinterest and Dow have tied workforce changes to AI but also noted how difficult it can be to know whether AI is the real cause of layoffs or simply the story companies want to tell Wall Street. Amazon itself told AP that AI was not the reason for the vast majority of its announced reductions, while Pinterest was much more explicit in tying cuts to an “AI-forward strategy.” AI is already affecting headcount decisions, investor expectations, and executive logic, even when it is not the only cause for concern.
That matters because this disruption is arriving in a labor market that is already softer and less prepared than many political leaders want to admit. The Associated Press reported this week that U.S. job openings fell to 6.9 million in February, gross hires dropped to 4.85 million, the hiring rate sank to 3.1% and resignations fell to their lowest level since August 2020. AP described the current labor market as “low-hire, low-fire” and noted growing worries that AI is taking over entry-level work while employers hold back on hiring until they better understand how they plan to use the technology. That is the context into which AI is being introduced, not a booming jobs market eager to retrain and absorb displaced workers, but a more hesitant one, in which early-career pathways are already narrowing. It is one thing to tell workers that technology will eventually create new roles. It is another to say that while opportunities to obtain stable work is already getting harder to find.
The opposing case also needs to be weighed carefully, because fear can be just as misleading as denial. Reuters reported in March that a European Central Bank blog post found firms in the euro zone making significant use of AI are, for now, more likely to expand hiring than reduce it and more likely to anticipate future employment growth than firms not investing in AI. Even so, Reuters noted that the longer-term outlook becomes murkier once AI begins to alter production processes more significantly. The same reporting has shown that in the United States, the sectors that have embraced AI most heavily have also seen more workforce contraction since 2022, with the Dallas Fed finding employment down 1% in the 10% of industries most exposed to AI even as total U.S. employment grew about 2.5%. At the same time, wages in those AI-exposed sectors continued to outpace the average, especially in roles requiring judgment and tacit knowledge.
That is the real picture—not a single sweeping collapse, but a growing divide. Some firms are still adding workers. Some sectors are shedding them. Some employees are seeing their bargaining power rise. Others are watching the path upward disappear.
That is precisely why regulation is needed. The world has experienced industrial change before, and neither automation nor technological disruption is new. Regulation is needed because the world confronting this transition is visibly unprepared for its speed, its asymmetry, and its distributive effects. The International Monetary Fund (IMF) said in January that nearly 40% of global jobs are exposed to AI-driven change and warned that concerns about job displacement and declining opportunities are becoming more acute. Its answer was not to leave the matter to market forces alone. It called instead for “proactive and comprehensive policymaking” to prepare workers and help ensure that the benefits are shared broadly. Governments need to take a serious look at this structural shock and recognize that the labor market will not adjust fairly on its own.
The weakness of the current regulatory landscape is not difficult to see. Europe has moved further than most. The European Commission says the EU AI Act entered into force on Aug. 1, 2024, with bans on certain practices and AI literacy obligations taking effect on Feb. 2, 2025, established rules for general-purpose AI models beginning on Aug. 2, 2025, and most of the framework becoming fully applicable on Aug. 2, 2026, with some requirements for high-risk products extending into 2027. The Council of Europe also adopted the first international legally binding AI treaty in 2024, establishing a framework meant to safeguard human rights, democracy, and the rule of law across the AI lifecycle. In the United States, by contrast, the most prominent federal tool remains the NIST AI Risk Management Framework, which the agency itself describes as voluntary. That is not meaningless, but it is also not a labor policy framework. The world is moving ahead with uneven and incomplete guardrails with one region building binding rules, one international body has laid down treaty principles, and one major power is still relying largely on voluntary guidance while companies move ahead much faster than governments.
What makes this gap especially hard to justify is that employment itself is already recognized as one of the places where AI can do outsized harm. The European Commission explicitly classifies AI tools for employment, worker management, and access to self-employment as high-risk uses. Its examples include CV-sorting software for recruitment. The same framework says high-risk systems should be subject to risk assessment and mitigation, quality requirements for data, traceability, documentation, clear information for deployers, human oversight, robustness, cybersecurity, and accuracy. Those are not radical demands. They are the minimum one would expect when algorithms are helping determine who gets interviewed, who gets promoted, who gets monitored, and who gets pushed out. If governments are willing to recognize that these systems can seriously affect fundamental rights, then the argument for regulating AI in employment is not futuristic. It is already built into the best-developed laws on the table.
The mistake in the United States has been to talk about AI regulation as though it were mainly a debate over speech, innovation, or existential safety, while giving less serious public attention to the labor market where the social effects will be felt first. If a company uses AI to screen applicants, workers should know. If an employer uses AI-based systems to evaluate performance, set schedules, or recommend terminations, workers should have notice, explanation rights, and access to human review. If firms save money by automating functions that once supported junior employees, there should be some obligation to invest in retraining rather than simply pocketing the margin and calling the result efficiency. The Council of Europe treaty already points in this direction by requiring transparency, oversight, accountability, and legal remedies for adverse impacts. The EU AI Act’s obligations around human oversight and traceability point in the same direction. These are not anti-business intrusions. They are guardrails against hidden decision-making in one of the most consequential parts of ordinary life.
None of this means regulation should be used as an excuse to stall progress. An unprepared world does not need empty bureaucracy any more than it needs blind faith in technology. Reuters’ reporting on the European Central Bank’s analysis is useful because it shows that, at least in the near term, AI may complement labor in some sectors rather than simply replace it. The International Monetary Fund has also emphasized that AI-related skills can boost wages and employment, while Reuters’ data suggests that workers whose roles depend on experience, judgment, and less easily automated knowledge are still seeing strong wage growth. Any serious regulatory approach has to make room for those benefits. The goal should not be to shut development down, but to restrain the most harmful uses, require accountability when AI shapes people’s livelihoods, and keep workers from becoming the default shock absorber for every technological leap.
The deeper issue is trust in the fairness of the system. People can live through enormous change when they believe the rules are just and the burdens are being shared honestly. But when they suspect the winners are shaping the rules while everyone else is told to adjust more quickly, trust begins to erode—not only in the technology itself, but in the institutions endorsing it. The International Monetary Fund’s 2026 analysis makes that risk hard to miss. It found that new skills tied to AI can lift wages and employment, but also deepen polarization, with the benefits flowing mostly to higher-skilled workers. Reuters’ reporting sharpens the picture further, showing that some AI-intensive industries are shedding jobs even as workers with harder-to-replace skills continue to see wage gains. That is how you end up with a smaller class of clear beneficiaries and a much larger group of workers being told their losses are unfortunate but necessary, temporary but somehow still evidence of progress. Some of those losses may eventually be offset. But societies are not held together by promises of future adjustment alone. They become unstable when too many people are asked to carry immediate pain for gains they are assured will arrive later.
And the workers most likely to absorb that pain first are often the ones with the least room to absorb it. The World Economic Forum has warned that employers expect workforce reductions where AI can automate tasks, while the Associated Press has reported growing concern that AI is already consuming entry-level work. If the first rung of the ladder begins to weaken, the damage does not remain isolated to one graduating class or one hiring cycle. It compounds over time. Fewer entry-level roles mean fewer places for people to learn, fewer future managers with accumulated experience, fewer midcareer workers with institutional memory, and greater pressure on younger employees to arrive fully formed for jobs that once taught them how to begin. A society that tells twenty-two-year-olds to simply reskill while shrinking the very roles that once helped them build skills is not overregulating. It is underregulating and pretending the market will somehow produce its own replacement ladder.
So yes, AI does need regulating in an unprepared world. Not because regulation can prevent disruption; it cannot. Not because governments are always prudent, because we know they are not. And not because every job loss linked to AI can be traced neatly to a model, a bot, or a redesigned workflow; many cannot. The case for regulation is more straightforward than that. Powerful systems are already shaping employment decisions. The labor market is already showing signs of strain. The pipeline for new skills is not keeping pace. And the rules now in place are too patchy and too voluntary for a transition of this scale. If AI is as consequential as its advocates insist, then it is too consequential to be governed mostly by quarterly earnings logic, corporate euphemism, and hindsight. A prepared world might have been able to treat AI as one more chapter in the story of productivity. This is not a prepared world. That is why the rules matter now.
