In April 2026, OpenAI published a report classifying US jobs by their exposure to AI, with a reassuring headline that the transition will be manageable. This paper works through the report’s own optimistic numbers and finds a different picture. Even read generously, the framework implies 7 million net US job losses over five years, comparable in scale to the 2008 financial crisis. Relaxing one assumption about physical work pushes the figure to 16.5 million. The demand-side drag implied by those displacement scenarios approaches Great Recession depth unless productivity gains arrive fast enough, are distributed broadly enough, and are matched by mitigation at the necessary scale. Independent modeling from Tufts University reaches similar magnitudes through different methods. The research infrastructure needed to see the transition clearly is technically feasible. The reason it is absent can be understood as a cumulative systems failure: AI companies, consultancies, academics, governments, the media, and the statistical instruments themselves each behave rationally. The aggregate outcome is that society cannot see what is happening to itself in time to respond.
Ewan Simpson (Thu,) studied this question.