A 4% AI jobs figure is a scenario. It is not evidence that 4% of jobs are gone.

What this paper did

Korinek, Jones, Sacher, Cotter and McCrory model how assumed AI capability paths could alter cognitive work, productivity, labor reallocation, wages and unemployment between 2026 and 2030. They illustrate modest, substantial and extreme scenarios rather than estimate an effect from firm or worker records.

The authors also survey U.S. adults about expectations. They say the median answers are consistent with the substantial scenario: GDP rises 8% and cognitive employment declines 4% by 2030.

Why the number is alarming—and why its label matters

A 4% decline in cognitive employment would be consequential. It is not a present-tense finding that AI has already eliminated 4% of jobs. The model’s result depends on what AI can do, how quickly firms deploy it, which tasks are automated, whether displaced workers move to other work and the other assumptions that define the scenario.

The paper does not assign a probability to the three paths on its accessible summary page. A survey respondent’s expectation does not supply that probability either.

The closest observed warning is still smaller and more specific

A Danish working paper follows firms that reported AI use in 2023 and finds employment growth 11% below non-adopters by late 2025 relative to prior trends, driven by lower recruitment and concentrated in smaller firms. That is an observed association after reported adoption. It does not prove AI caused the gap, predict other countries or count a broad layoff wave.

Broad U.S. job-posting research, California exposure-linked unemployment claims and Nordic records put further limits on a single collapse story. They do not erase the Danish hiring warning because they use different places, treatments and outcomes.

What changes in the cumulative answer

The pool now has a live example of a precise future employment number that must not be recast as observed displacement. The current evidence still points to possible concentrated hiring restraint after adoption and exposure-linked early-career risk, not a measured economy-wide job-loss verdict. Read any forecast with its horizon, assumptions and named outcome intact.