A Sequential Foresight–Delphi Protocol for Diagnosing Anticipatory Legitimacy: Futures Wheel, Causal Layered Analysis, and Delphi Applied to Artificial Intelligence‐Powered Digital Government in Vietnam

Futures and foresight research offers mature methods for generating consequences, interpreting their meaning, and converging expert judgment, but rarely integrates them into fully specified protocols that diagnose an emerging technology's legitimacy risks before they materialize. This study develops and applies a Sequential Foresight–Delphi Protocol linking Futures Wheel consequence mapping, Causal Layered Analysis (CLA), and Delphi appraisal through prespecified transition rules, ex ante consensus criteria, a stability‐based stopping rule, and retention of dissensus. The protocol diagnoses anticipatory legitimacy, the prospective justifiability of AI‐enabled public authority, across consequential visibility, institutional responsiveness, and narrative alignment. It was applied in Vietnam, a critical illustrative case of compressed digital transformation, with a Quadruple Helix panel (26 invited; 20 completing every round). The panel anticipated efficiency gains cascading into accountability gaps, profiling, exclusion, and faceless administration; CLA traced the risks to a technocratic worldview and participant‐supplied images condensed as the algorithmic mandarin. Across two original Delphi rounds (urgency, feasibility) and two supplementary rounds (legitimacy‐enhancing potential), the panel converged on impact assessment, public registers, human review, and sandboxes, whereas contestability‐by‐design and independent audit reached consensus on urgency and legitimacy‐enhancing potential (for audit, without meeting the stability criterion) yet did not converge on feasibility, retaining the portfolio's two lowest feasibility ratings. We advance this asymmetry as an open–closed proposition for comparative testing. All claims concern expert‐anticipated consequences and judgments; no causal, predictive, or statistically generalizable claims are made.