Job Retraining Programs Show Modest Gains, Study Finds

Worker retraining programs, long touted as a solution to labor market disruption, produce modest benefits at best, according to a new review published by Anthropic’s Economic Research team. The analysis, co-authored by David Roodman and Maxim Massenkoff, evaluates 56 randomized U.S. studies alongside European experimental evidence, offering a sobering look at the challenges ahead as AI reshapes global employment.

The findings reveal that retraining programs increase employment by just two to three percentage points and boost annual earnings by approximately $1,000 per participant. However, these small gains come at a steep cost—roughly $13,000 per trainee. Government budgets recoup over half of the investment through additional tax revenue and reduced benefit payments, making the programs close to break-even overall.

Sector-specific programs tied to high-demand industries stand out, delivering significantly better results. These initiatives, which partner with employers and place participants directly into roles, achieve much higher employment and earnings gains. However, attempts to replicate such successes have frequently failed, underscoring scalability challenges. “If AI displaces workers at scale,” the authors warn, “existing retraining programs would likely fall short.”

The research arrives amid growing interest in active labor market policies. The OECD’s Employment Outlook 2026 emphasized the need for targeted retraining and reskilling efforts, particularly in communities hit hardest by deindustrialization. Similarly, a 2026 ILO-World Bank report highlighted how program design, employer links, and certification are critical to maximizing impact. The European Commission has also pointed to the importance of integrating training with job placement services and employer partnerships for better outcomes.

Anthropic’s report strengthens these conclusions with meta-analysis data, but also raises tough questions about whether retraining can scale effectively in the face of rapid AI-driven changes. The authors advocate for immediate investment in piloting and evaluating the most promising models, calling on policymakers to prioritize rigorous testing and expansion of successful sector-specific initiatives. Anthropic’s Economic Futures Research Fund is designed to support such efforts.

For policymakers and stakeholders, the message is clear: retraining isn’t a one-size-fits-all solution. Programs work best when aligned with real labor market demand, embedded with employer participation, and delivered alongside job placement support. Without these elements, large-scale AI displacement could outpace the ability of retraining systems to mitigate its effects.

As industries brace for automation’s impact, the window for proactive planning is narrowing. Governments and organizations must act now to scale effective retraining models or risk leaving displaced workers with inadequate support.

Image source: Shutterstock