AI labs' job-loss plans won't say what the public gets

Frontier AI labs have begun publishing elaborate economic-policy plans for a future where AI sharply reduces labor demand, but they repeatedly avoid the central accounting question: how large a public claim on capital should replace lost labor income. Anthropic’s three-tier framework ties incremental measures - wage insurance, retraining, extended unemployment benefits - to rising unemployment and reserves taxes, sovereign wealth funds, or basic income for a final, catastrophic tier it admits it cannot map. Across seventeen policy statements addressing distribution, fourteen refuse to specify what share of AI-created wealth should go to the public. That pattern is what the author calls “preparedness theater”: public displays of planning that omit the basic redistribution arithmetic while building research funds, fellowships, advisory bodies and reputational signaling (Ben Bernanke’s inclusion among them) to show seriousness without committing to concrete claims.

The substantive point is straightforward: if AI shifts income from labor to capital, transfers that once drew on wages must instead draw on capital returns, so the key policy is how to secure public ownership or dividends from capital - social-wealth funds or universal basic capital - rather than only labor-market triage. Some lab papers, like DeepMind’s, propose ownership-focused backstops but use opaque scoring and simulation methods that leave implementation-size and timing unspecified. Many advocate waiting for clear data before large interventions, a posture that risks acting only after mass income loss. The debate reduces to who will set the share: private firms, technocrats, or democratic/public institutions wielding capital claims.