Human Policy Documents, written primarily for human readers, are widely used in current AI governance. They describe principles that AI should follow, such as respecting human rights, promoting fairness, ensuring transparency, operating safely, and protecting privacy. These principles are an important starting point for defining the values and requirements of AI governance. The problem is that these documents assume human interpretation.

Humans can read a policy while considering its purpose, background, and social context, and determine which principles apply in a particular situation. Giving the same document directly to AI, however, does not necessarily produce the same conclusion every time.

For example, three AI systems may be Reading the same policy yet reach different decisions: Allow, Review, or Block. This does not mean that the AI systems are malfunctioning. Policies written in natural language contain expressions, conditions, exceptions, priorities, and scopes that require interpretation. If that interpretation itself is delegated to probabilistic AI, governance decisions may also become probabilistic.

This creates an important boundary: Policy document and Policy execution are not the same thing.

Even when laws, regulations, organizational policies, and ethical principles exist, their existence alone does not mechanically determine which rule an AI system should apply, in which situation, or under whose authority.

Human-readable policy documents therefore do not need to be eliminated. What is required is a translation layer that preserves their institutional meaning while transforming their requirements into governance structures that AI systems can use.

In I2EA, the Translation Hub performs this function by translating institutional requirements into forms that systems can process, with the results represented as structured information such as Governance Tags. At the execution stage, rather than allowing AI to freely reinterpret the policy each time, this governance information is connected to the Execution Boundary.

Human-readable policy is necessary, but it is not executable governance.

The challenge is therefore to preserve institutions designed for human understanding while transforming their requirements into governance structures that AI systems can operationally use. This is one of the central problems addressed by Institution-to-Execution.