Policy and Investment Recommendations for Trustworthy AI

This is the twin document to the ethics guidelines, and the one that has aged worse of the two. One set out the principles, the other the industrial policy, and it is the industrial policy that failed.

The June 2019 recommendations correctly identified the levers: computing capacity, data access, skills, funding from seed stage through to growth. The error was not in the list but in the order of magnitude. The amounts envisaged were calibrated for a research economy, not for an infrastructure race. The comparison is a cruel one: France alone announced 109 billion euros in investment in February 2025, and the annual capital expenditure of the major American players now exceeds what Europe was contemplating for an entire decade.

The document also rightly identifies Europe’s dependence on non-European infrastructure. Seven years on, that dependence has only grown, including in investments presented as sovereign but mostly financed and operated from abroad. A set of recommendations that gets the diagnosis right without bearing on actual allocation decisions remains, in the end, an exercise in analysis.