Can we separate reliability, safety, and trust into distinct mechanisms?

The same paper's Method and Conclusion sections break "Reliable, Safe & Trustworthy (RST)" into three distinct mechanisms rather than treating trustworthiness as one undifferentiated quality: a system "will be (1) reliable because of sound technical practices, (2) safe because of open management strategies that build cultures of safety, and (3) trustworthy because of well-designed independent oversight structures." Each term is tied to a different lever — reliability to engineering practice, safety to management culture, trust to governance structure — so that improving one does not automatically improve the others.

For trust specifically, the paper leans on Fukuyama's (1995) definition of social trust as arising "within a community of regular, honest, and cooperative behavior, based on commonly shared norms," and argues this community-based notion of trust is earned by relying on "respected independent oversight structures," citing the author's own earlier work (Shneiderman, 2000, 2016) on what such bodies look like. Reliability draws on technical practices that support "human responsibility," "fairness," and "explainability." The paper ties the triad to concrete design outcomes: "human responsibility for mistakes is another powerful driver of design advancements such as the inclusion of detailed audit trails and consistent informative feedback about machine state," which in turn speeds "retraining of users and redesign of systems."

This gives a structural answer to a question the vault's workplace-agent research otherwise raises only empirically. What UX principles do workplace users want in AI agents? finds users rank reliability and safety among their top design principles but does not say what organizationally produces them; this paper supplies that account — reliability from technical practice, safety from management culture — as separable prerequisites rather than properties that simply emerge from a well-designed agent. It also sharpens the oversight language in Does granting agents more autonomy undermine human oversight?: where that paper treats the overseer's presence and personal capacity as the oversight problem, this paper's trust leg locates the oversight structure itself — independent, audited, external to the individual user — as the trust-bearing mechanism, a different unit of analysis (institution rather than individual reviewer).

The excerpt states the RST triad as a framework, not a tested model: it offers no measurement of reliability, safety culture, or oversight independence, and says outright that "actionable guidelines to achieve these goals are beyond this paper." The audit-trail and retraining claims are asserted as design logic, not demonstrated with deployment data. What follows at that strength: the triad is useful for separating "is this system trustworthy" into three independently auditable questions, but answering any one of them still requires the measurement work the paper itself defers to future research.