Cognitive Operating Architecture: The Missing Layer in AI Governance

Contemporary approaches to AI governance concentrate on model safety, alignment, ethics and regulatory compliance. These are necessary, and they rest on an assumption that is becoming invalid: that institutions retain stable authority, accountability and decision ownership as cognition is progressively delegated to machine systems. This paper identifies and defines the cognitive operating architecture (COA): the institutional layer governing the conditions under which machine-generated judgements are relied upon, how authority accumulates through repeated use, and how accountability persists across system updates, organisational turnover and hybrid human--machine decisions. Failures commonly attributed to "AI risk" are frequently failures of operating architecture rather than of model behaviour or ethical intent. By distinguishing the COA from technical AI safety, from ethics frameworks and from formal governance structures, the paper reframes AI governance as an architectural problem. The absence of the layer produces predictable pathologies: institutionalised automation bias, ritualised oversight, authority laundering, responsibility diffusion and effectively irreversible dependence on opaque cognitive systems. The paper situates the COA alongside the institutional operating architecture, and specifies the mechanism layer it requires but does not itself provide. It does not propose solutions to AI governance; it establishes the architectural precondition that makes governing delegated cognition possible.