The anxiety over artificial intelligence's impact on employment has become a fixture of tech discourse, yet it rests on remarkably shaky ground. At Anthropic, one of the field's leading AI safety organizations, a team of societal impacts researchers is attempting to change that by building the first comprehensive dataset on how AI actually affects workers. The project, still in early stages, aims to move beyond anecdotal horror stories and venture-capitalist speculation toward empirical evidence. However, the researchers face a daunting challenge: the tech industry has historically resisted producing transparent employment data, and no government agency currently mandates standardized AI impact reporting. The Anthropic team is working with limited funding and unclear timelines, raising questions about whether independent research can fill a gap that arguably demands regulatory oversight.

Real-world evidence of AI's effects is scattered and incomplete. Consider Mike McClary, who runs a small online outdoor brand selling specialized flashlights and gear. In recent years, McClary has watched AI-powered recommendation algorithms and competitor analysis tools fundamentally reshape his product development decisions. What once relied on customer intuition now depends on algorithmic predictions of market demand. While McClary's story illustrates how AI is automating aspects of entrepreneurial decision-making, it also highlights the broader problem: there is no systematic way to quantify how many small business owners face similar disruptions, or whether they're being pushed toward riskier, algorithm-optimized product lines. Economic researchers attempting to measure AI's employment effects have produced conflicting estimates, ranging from negligible job losses to wholesale displacement of 300 million positions globally by 2030, a variance that underscores the absence of reliable baseline data.

The stakes are prompting frustration among policymakers. Several members of Congress have privately expressed concern that they lack the empirical foundation needed to craft effective regulation or support programs. Building reliable datasets would require mandatory disclosure from major AI companies about their automation decisions, longitudinal employment surveys tracking affected workers across sectors, and regular audits of AI systems deployed in hiring, customer service, and data-processing roles—precisely where job losses are most commonly reported. The European Union's AI Act includes nascent reporting requirements, but the United States has no equivalent framework. Until such infrastructure exists, policy responses will remain reactive, addressing crises after they've already transformed labor markets. The Anthropic researchers estimate that a robust national employment-impact database could be operational within 18 to 24 months if properly funded and mandated, but without government backing or industry cooperation, the initiative risks becoming another incremental research effort in an ocean of uncertainty.