KERALA RISING · FROM EXPORTING PEOPLE TO EXPORTING IDEAS safeguards are real, but they are paper, and paper has lost to power before. The technology may not be as ready, or as cheap, as claimed The economic case depends on verification being accurate and nearly free. In a clean benchmark it is. In the field, the camera angle is wrong, the lighting is poor, the categories blur, the model is confidently mistaken, and the cost of escalating the hard cases to human review or better models climbs. If the realworld error rate is high enough to require heavy human checking, the cost advantage that makes the whole scheme affordable erodes, and the institution becomes just another expensive programme that happens to use AI. The published pricing is real; the assumption that the civic case stays at the cheap end of it is a forecast, and forecasts about technology are wrong all the time. It may work, and still not matter Suppose every objection above is overcome. The loop transfers, fraud is contained, equity holds, the institution stays honest, the technology stays cheap. It is still possible that the gains are real but marginal, a somewhat cleaner corridor, a slightly safer road, that do not justify the institutional apparatus built to produce them, and that the same money spent conventionally would have done as 310