Last Updated on February 5, 2026 by Chicago Policy Review Staff
As artificial intelligence models start performing wider ranges of tasks, companies are increasingly claiming to be cutting jobs due to AI. In May, Dario Amodei, the CEO of AI lab Anthropic predicted AI will cause unemployment as high as 20% within five years. In September, Salesforce cut 4,000 employees, saying they were no longer necessary due to AI. However, an unanswered question hangs in the air: What does the data say about how AI affects workers? A bipartisan bill aims to get to the bottom of this question by requiring companies to report AI-related employment changes.
Automation in context
The idea that AI is causing layoffs has been getting a lot of buzz. But economists disagree about whether AI will eliminate more jobs than it creates over the long term — and even whether it is displacing workers yet.
Some studies have found that, since the release of ChatGPT in 2022, early-career workers in roles with more AI-replaceable tasks have tended to see more unemployment. For example, three economists from Stanford University found that early-career workers in AI-exposed occupations experienced a 16 percent relative employment decline.
But other work has found that AI is not having much of an effect on the labor market yet, such as a recent paper that linked administrative labor records to survey responses from 25,000 workers across 11 of the most AI-exposed occupations in Denmark. Rather than just looking at which occupations were most AI-exposed, the authors, Anders Humlum and Emilie Vestergaard, examined the extent to which firms were actually adopting AI and found that adoption was not correlated with reduced labor demand: “we estimate precise null effects on earnings and recorded hours at both the worker and workplace levels, ruling out effects larger than 2%.”
Although automation renders certain tasks obsolete, experts expect that it will also create new tasks and increase the demand for labor in unautomated tasks. However, it is hard to predict which of these effects will be larger in the long run. Some suggest that AI’s ability to adapt to new domains will allow it to automate even the tasks that it creates.
As with many other economists, history is a cause for optimism for Humlum, an assistant professor of economics and the Fujimori/Mou Faculty Scholar at the University of Chicago Booth School of Business. He points out that labor’s share of total economic output has “remained remarkably stable” since the Industrial Revolution despite waves of technologies changing the nature of work. For example, hundreds of years ago, most of the workforce was dedicated to agriculture, but now only 1.2 percent of the U.S. population works on farms. “It is due to technology, advanced harvesters and so forth, but that has not limited humans in their ability to provide value,” says Humlum. “In some sense it has just augmented us in the sense that we don’t need to spend all our time growing food, and we can think about other things we can create that maybe have more value.”
AI-Related Job Impacts Clarity Act
Senators Mark Warner (D-VA) and Josh Hawley (R-MO) have both expressed concern about AI’s potential employment effects. Warner mentioned worries about entry-level roles specifically, saying in a CNBC interview: “If we eliminate that front end of the pipeline, how are people ever going to get to that mid-career spot?” In December, he introduced legislation with Representative Raja Krishnamoorthi (D-IL), creating a tax incentive for employers to invest in worker re-training. Hawley, for his part, released a broad framework for AI legislation with Senator Richard Blumenthal (D-CT) in September 2023 and since then has sponsored bipartisan AI-related bills on topics from child protection to evaluation of advanced models.
On Nov. 5, Warner and Hawley introduced the AI-Related Job Impacts Clarity Act, which would require covered employers to submit quarterly reports to the Department of Labor (DOL) detailing AI-related employment changes. Companies would be required to disclose the number of employment actions, including layoffs, new hires, retraining, and unfilled positions, attributable “substantially” to AI.
Based on these disclosures, the department would compile quarterly and annual reports, publish them on the Bureau of Labor Statistics website, and submit summaries to Congress. Twice a year, DOL would also analyze AI-related net employment trends.
The act would cover all federal agencies and publicly-traded companies. It leaves it up to the Secretary of Labor to determine the extent to which non-publicly-traded companies would also have to report information.
The need for data
Uncertainty about how AI will alter the future of work is one reason to seek better data: This information can enable more evidence-based discussions and policies. Another is that, even if automation creates more jobs than it eliminates, it may displace many workers from some sectors. For example, automation may have eliminated 400,000 manufacturing jobs between 1990 and 2007, famously causing strain in communities where these jobs were concentrated (Acemoglu & Restrepo, 2020). A better understanding of where these shifts are happening can inform policies that help workers adjust.
“We also need to do more work on mapping out the different pathways for workers that both speak to them in the sense that they are something they could see themselves doing but also provide them a way back into the labor market without having repeated displacement from AI,” Humlum says.
In a September letter to DOL, several prominent economists, including four Nobel laureates, called for improved data collection on AI’s labor market impacts to help policymakers plan for future scenarios and know which policies are most effective. Areas in which existing data could be improved, advocates argue, include increasing response rates and sample sizes in government surveys and tracking the task content of jobs more frequently to see how it changes over time. Humlum, for his part, suggests asking more specific questions about AI use, looking at which industries use AI the most, and tracking how much money is spent on AI in the workplace.
Benefits and burdens
Critics have said this bill may create a regulatory burden for employers or cause negative publicity for companies that replace jobs with AI. The text of the bill does not make clear whether DOL’s public summaries will mention individual employers or only include aggregate data.
Another question the bill leaves open is, “What counts as an AI-caused employment change?” This is more ambiguous than it may sound. Some experts argue that companies have been using AI as an excuse for layoffs that actually happen for more mundane reasons. Even if employers try to report accurately, they may make inconsistent judgements, given that multiple factors often contribute to employment changes.
“I don’t want it to be our only measure of job loss due to AI because there’s a bias because AI is very salient, […] but there are a lot of things going on in the economy right now,” Humlum says.
Rather than relying on employers’ determinations of causation, researchers could infer the effects of AI by looking at levels of AI adoption and comparing them to employment patterns — as in Humlum and Vestergaard’s paper. This would allow for more objective causal inference about which occupations are being impacted, rather than relying on potentially inconsistent judgments. In addition to the data listed in the bill, the act also allows DOL to require “any other information related to artificial intelligence-related job impacts, as determined appropriate by the Secretary,” which could facilitate this analysis.
But despite the limitations of self-reported causality, Humlum says these responses could complement independent inference by serving as an “initial flag” that layoffs are happening, which may be more timely than retrospective analysis.
Private alternatives
Humlum says that it can be hard for academics to obtain good data on AI adoption due to the cost of running large surveys. AI labs hold most of the high-quality data, since they are able to track this information from their users.
“Maybe the government should require [AI companies] to disclose usage rates by industries and firm characteristics,” Humlum says. “I think that would be quite informative, and […] it would not be very costly in the sense that the data is already being collected. It would not require firms to fill out all these forms like this new bill would require them to do, which would be more burdensome in terms of the administrative cost.”
Anthropic recently released data on how their models are being used in the economy, which has been used in automation research already. For example, the study by Stanford researchers from last year used this data to evaluate which occupations used AI the most. This study also used data from payroll service firm ADP to measure job-related changes. In addition to these examples, a few other private firms also gather and analyze data on this issue to inform business decisions.
Many questions exist about how AI will influence the economy — whether it will create more jobs than it renders obsolete, whether its current effects are severe or overhyped, whether it will create skyrocketing growth or be closer to business as usual — and the AI-Related Job Impacts Clarity Act may not answer all of these. But despite the limitations of the self-reported data that the bill calls for, it could supplement independent analyses and provide a stepping stone for future evidence-informed policy responses.