Beyond Trustworthy AI: Unstable Trust, Mental Health AI, and Proof Infrastructures in Healthcare Governance

Healthcare AI governance increasingly relies on the paradigm of "trustworthy AI," which assumes that explainability, transparency, and compliance mechanisms can stabilize trust in automated systems. This article challenges that assumption through a socio-legal analysis of healthcare AI governance based on findings from 28 interviews conducted with French health-data governance actors. The interviews revealed fluctuating perceptions of AI depending on institutional position, operational exposure, and responsibility allocation. Trust appeared conditional upon supervision, secure infrastructures, traceability, and accountability rather than solely upon technical explainability. These tensions become particularly visible in mental-health AI, where systems increasingly rely on multimodal behavioral and contextual data to infer psychiatric or psychological states under conditions of persistent uncertainty. The article argues that explainability alone struggles to stabilize trust in such environments because uncertainty concerns both the interpretation of AI systems and the meaning of the underlying data. It consequently proposes a shift from stable trust assumptions toward proof-oriented governance centered on traceability, auditability, contextual continuity, and distributed accountability. Within this perspective, blockchain-based architectures such as Galéon are examined as limited proof infrastructures supporting evidentiary continuity rather than as autonomous "trust machines." The paper concludes that healthcare AI governance should focus less on producing durable trust than on preserving the conditions under which uncertainty remains visible, verifiable, and governable.