Earned, Not Manufactured: The Regulatory Gap in AI "Trust" that Behavioral Science Can Close

As chatbots become increasingly capable of engaging in natural, human-like conversation, understanding how users develop trust in these systems is becoming increasingly important. Regulatory and policy frameworks tend to define trust in normative terms such as transparency, accountability and human oversight. However, the trust users place in chatbots often emerges through different mechanisms: cognitive biases, anthropomorphic cues, and interactional design choices that can generate trust independently of whether any normative standards are met. We argue that conflating these two phenomena under the shared ``trust'' label obscures a critical distinction between psychological trust formation and normative trustworthiness, and leaves a gap neither body of work can address alone. Normative frameworks for trustworthy AI are necessary, but they need grounding in concrete behavioral evidence about what actually drives calibrated user trust, which can be manufactured by exploiting our cognitive biases. This paper calls for a dual framework separating these two phenomena and for an empirical research agenda to identify the behavioral drivers of earned trust to complement and strengthen existing regulatory frameworks.