Institutional AI safety in 2026 is producing answers to questions the technical community stopped treating as primary. Governor Shapiro's roundtable [4] and the Trump administration's legislative framework [5] both represent policy operating at the behavioral layer — what models say, whether children are protected, whether speech is preserved. The structural layer — who owns the compute, who captures the gains, who controls the training data — is not the subject of any of these initiatives. That gap is not an oversight; it reflects the genuine difficulty of legislating infrastructure concentration.…