Corporate ethical and responsible AI (CERAi) and AI trustworthiness: a lifecycle perspective and empirical insights

Abstract

Purpose This study develops and empirically validates a Corporate Ethical and Responsible AI (CERAi) lifecycle framework. The framework comprises three interdependent stages: design and development, testing and training, and deployment and usage. The study aims to move the research beyond abstract principles and policy analyses toward empirically grounded, lifecycle-integrated artificial intelligence AI governance theory. It positions ethical AI and responsible AI governance as the two constitutive components of the CERAi construct, and AI trustworthiness as a conceptually and empirically distinct downstream outcome anchored in stage-specific organizational governance mechanisms. The CERAi framework is scoped broadly to encompass governance of both conventional/traditional machine learning models and autonomous AI agents operating within organizational settings. Design/methodology/approach We administered a large-scale survey to 500 attendees at a national AI conference in the United States in early 2026, yielding 143 useable responses from AI professionals across healthcare, financial services, manufacturing, retail and energy sectors. We developed and validated multi-item scales for each CERAi lifecycle stage using established ethical AI and AI governance frameworks. Structural equation modeling was employed to assess the measurement model and to test both direct path hypotheses and mediation hypotheses. Findings Both CERAIi design and development and CERAi deployment and usage have significant positive direct effects on AI trustworthiness, whereas CERAi testing and training do not. Our results also indicate a significant serial mediation pathway in which CERAi deployment and usage enhance AI trustworthiness through testing and training, followed by design and development. Conversely, testing and training alone do not mediate the relationship between deployment and usage and AI trustworthiness, indicating that design-stage ethics serve as an intervening (mediating) mechanism through which governance activity enacted at later lifecycle stages is associated with perceived AI trustworthiness. This is consistent with testing and training's association with trustworthiness operating indirectly, mediated by the ethical architecture established during the design and development stage. Originality/value Findings of this study demonstrate that AI trustworthiness reflects differentiated perceived salience across AI lifecycle stages, whereby stakeholders appear to weight visible, deployment-stage governance more heavily than upstream design and testing stage governance in forming trustworthiness judgments, rather than reflecting a definitive validation that governance effects accumulate uniformly across all stages in isolation. This study makes a distinctive contribution by providing empirical evidence for a governance-sequencing mechanism, whereby design-stage ethics function as a mediating pathway shaping the effectiveness of governance activity at later lifecycle stages. To the best of our knowledge, such a mediated, stage-sequential mechanism has not previously been demonstrated using primary data within a comprehensive, three-stage CERAi lifecycle framework linking organizational ethical and responsible governance practices to AI trustworthiness outcomes.