Abstract <p>The delegation of human acts to Artificial Intelligence (AI) systems challenges the adequacy of current ethics and governance frameworks. This challenge arises because these frameworks typically operate within a technological instrumentalist paradigm, framing AI as a value-neutral tool. This perspective evaluates actions through the Standard Instrumentalist Model of Responsibility (SIMR). However, critics argue that the SIMR is inadequate for evaluating AI-mediated acts because it introduces a teleological inversion into those actions. This inversion produces an ontological shift that engenders a normative vacuum, wherein the search for a linear path to identify human intent throughout the AI-mediated process diffuses. To address this inadequacy, this paper defends the Epistemic Uniqueness Thesis (EUT). The thesis challenges both the instrumentalist framework and its corresponding model of responsibility attribution by arguing that AI-mediated acts generate a "responsibility gap." This gap is the result of two core features: agency diffusion and technical/teleological opacity. To operationalize these concepts, the paper introduces two formal mechanisms: the Praxis-Capture Criterion (PCC) and the Agency-Diffusion Function (ADF). These mechanisms illustrate how the systemic complexity of AI disperses agency below the threshold of human perceptibility, making it impossible to pinpoint a responsible moral agent within the causal chain. Ultimately, the diffusion of agency necessitated by the teleological inversion of AI-mediated acts underscores the need to reevaluate current AI governance—specifically by rejecting the Value Neutrality Thesis, which falsely claims that AI systems are value-neutral and that humans are the sole carriers of values.</p>