Governing AI-Assisted Board Decisions: A Dual-Lock Model for Information Independence and Accountability

Purpose: This study develops a governance architecture for board-governed strategic decisions in which artificial intelligence (AI) provides common advice while legal and organizational decision ownership remains human. It examines two linked risks: erosion of independent human information and distortion of delegation through outcome-dependent credit and blame. Methodology/Approach: A theoretical decision-economic model is used to formalize firm-optimal and executive-optimal delegation, an endogenous attribution wedge, common-AI-induced error covariance, and boundary conditions for pre-AI information recording. The analysis treats governance controls as ex-ante commitments rather than ex-post corrections. Findings: Common exposure to an imperfect AI signal can increase correlation among revised human judgments and reduce the effective number of independent checks. Separately, private attribution incentives can move delegation away from the firm-optimal benchmark. The proposed Information Lock records designated human assessments before common AI exposure, while the Accountability Lock fixes effective decision rights and attribution rules before outcomes are known. The model also identifies limits: perfectly reconstructible post-AI judgments reduce the Information Lock mainly to an auditability device, while residual career downside can shift distortion from over-delegation toward rejection or delay of positive-value decisions. Conclusion: The contribution is a unified, selective governance architecture linking information independence and outcome-invariant accountability. It provides boards and senior executives with an implementable framework for preserving informational diversity, clarifying decision ownership, and distinguishing poor process from unfavorable stochastic outcomes in AI-assisted strategic decisions.