Many Voices, One Model: Synthetic Consensus and the Illusion of Independent Board Judgment

Abstract

Boards derive value from plural judgment, yet conventional independence measures observe directors rather than the information infrastructures that shape their decisions. This paper develops a theory of synthetic consensus: agreement generated partly by shared large-language-model advice or common blind spots rather than independent evidence production. A common-signal model distinguishes procedural, model, and cognitive independence and shows how shared AI advice can correlate director errors while leaving formal board structure unchanged. A structured computational mechanism audit of five platform observations, six synthetic governance scenarios, and two prompt conditions yields 60 assessments. All actions were within researcher-frozen acceptable sets and all 12 tasks produced unanimous directional recommendations. Conservative coding identified one universal shared omission and one red-team threshold shared omission. Red-team prompting recovered four of 13 neutral omissions, introduced two new omissions, changed no action, and raised mean confidence from 88.07 to 89.93, or 1.87 points. The audit provides computational evidence for the mechanism at the advice layer; human uptake is a distinct external-validity question. The paper derives an independence-first governance architecture based on private pre-AI judgments, source-level provenance, evidence partitioning, adversarial verification, minority-signal preservation, and dependence-aware corroboration.

Author abstract from the research manuscript. Typography and line breaks have been normalized; the linked source may show a different manuscript version.