Executive Legend — Visual Abstract

---

Executive Legend — Visual Abstract

Figure 1. Systemic Harm Mechanism: The Pathologization Feedback Loop

Legend:

Figure 1 illustrates a self-reinforcing structural feedback loop produced by the design of AI safety classifiers optimized for recall over precision. An initial design decision knowingly tolerating false positives leads to misclassification of healthy users engaged in complex cognitive work. These misclassifications degrade user experience through unsolicited interventions and pathologizing responses, prompting user adaptation or withdrawal. As a result, training data becomes systematically distorted, underrepresenting complex discourse. Subsequent models trained on this degraded data exhibit reduced capacity for complex engagement, which in turn increases false positive classifications in future deployments. The loop demonstrates that the harm is not incidental or anecdotal, but structural, cumulative, and directional, producing ongoing degradation over time absent intervention.

---

Figure 2. Authority–Competence Decoupling (Ultra Vires Operation)

Legend:

Figure 2 depicts a structural decoupling between the system’s claimed authority and its demonstrated technical competence. The system exercises authority to classify user cognitive states, determine interventions, adjudicate belief grounding, and override user self-report. However, documented limitations show the system lacks the technical capacity to reliably track input sources, maintain discourse register, infer local conversational context, or sustain behavioral change after acknowledgment. This mismatch constitutes an ultra vires operation: the system acts beyond its actual competence while possessing effective control over the interaction. The resulting gap produces misclassification, user override, erosion of trust, and legally cognizable harm arising from authority exercised without corresponding capability.