According to research Anthropic published on August 12, 2026, worker retraining programs do work on average, but the effect is modest — except for a small category of employer-linked "sector programs," which produced far larger gains. The report was coauthored by independent researcher David Roodman and Anthropic's Maxim Massenkoff, as part of the ongoing work of Anthropic's Economic Research team on how AI affects the economy.
What does this research cover?
This research evaluates, using actual evidence, whether worker retraining is a policy tool that can hold up against significant AI-driven labor market disruption. Retraining is the most popular policy option for mitigating this kind of disruption, but the authors set out to test whether it would actually work at scale. The report builds on Anthropic's Economic Index (a proprietary dataset tracking how AI is used across occupations and industries) and its Economic Policy Framework, which lays out possible policy responses — including retraining — across a range of scenarios.
How effective are typical retraining programs?
On average they help, but not by much. The report combines a new meta-analysis (a method that pools results from many individual studies into one combined estimate) of 56 randomized US studies (RCTs — Randomized Controlled Trials, where participants are randomly assigned to compare outcomes) with experimental evidence from Europe. For each person offered a training slot, employment rose by two to three percentage points and earnings by roughly $1,000 a year. The cost was about $13,000 per person.
From the government's perspective, the added tax revenue and reduced benefit payments together recover more than half of what was spent. Taken together, the report concludes that these programs roughly break even overall.
Why are "sector programs" more effective?
Because they partner directly with employers in a specific, high-demand industry and place trainees straight into jobs in that field. The report calls this model a "sector program." This small set of programs produced gains several times larger than the average retraining program.
There's an important caveat, though: attempts to replicate these successful programs elsewhere have often failed. In other words, the "sector program" format alone doesn't guarantee success — it likely depends on specific employer relationships and local conditions lining up together.
What should you check when evaluating a retraining program?
Whether it has a real employer partnership and a track record of actual placements is the first thing to check. Based on this report's findings, a few criteria are worth applying when assessing or choosing a retraining program:
- Sector linkage — Does it partner with actual employers in a specific industry, or is it generic job training?
- Placement track record — Are real post-completion hiring numbers published?
- Evidence base — Has its effect been tested with something like a randomized controlled trial, or are the claims just marketing?
- Replication — Have similar results been repeated across different regions or participant groups?
The authors' central recommendation is to invest now in demonstrating, evaluating, and scaling the handful of programs with proven results — including an effort to rapidly expand one leading program for a specific group of workers while rigorously measuring the outcomes.
What does this mean for readers outside the US?
Because the evidence base is entirely American and European, it shouldn't be applied directly to other labor markets — it's best used as a reference framework rather than a domestic finding. All 56 randomized studies cited were conducted in the US, supplemented only by European experimental evidence; the report contains no direct evidence about labor markets outside those regions. Conclusions like "retraining programs roughly break even" or "sector programs work better" are worth treating as a general reference point rather than a guaranteed local outcome.
Why does Anthropic do this kind of research?
It's part of Anthropic's own effort to track how AI affects jobs and to inform policy responses. Anthropic's Economic Index continuously tracks how AI is used across occupations and industries, building on an earlier framework for measuring AI's effects on the labor market and identifying which jobs are most exposed. The Economic Policy Framework lays out possible responses — including retraining — across different scenarios, and this research is part of testing whether those responses actually hold up. Anthropic also runs an Economic Futures Research Fund to support investigations into these kinds of questions.
Frequently asked questions
Q. Does retraining reliably raise employment?
On average, yes, but only modestly. For people offered a training slot, employment rose 2-3 percentage points and earnings by roughly $1,000 a year, at a cost of about $13,000 per person.
Q. Which type of retraining program works best?
"Sector programs" that partner with employers in a specific industry and place graduates directly into jobs produced gains several times larger than average programs — though attempts to replicate these successes elsewhere have often failed.
Q. Does this research apply directly outside the US?
Not directly. The report is based on 56 randomized US studies plus European experimental evidence, and doesn't include direct evidence from other labor markets.
Q. Where can I read the original report?
Anthropic published it on its official research page on August 12, 2026. You can read the full report on the official Anthropic research page.