Technical–ethical–governance circuit–breaker framework for safe and sustainable AI governance

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The growing use of large language models and autonomous AI systems in high-risk settings has created safety risks that cannot be addressed through technical safeguards alone. Existing approaches often treat technical failure control, ethical oversight, and institutional governance as separate domains, making coordinated intervention difficult when risks escalate across these boundaries. This study proposes the technical–ethical–governance (TEG) framework as an integrated circuit-breaker architecture for managing risks in high-risk AI systems. The framework combines technical, ethical, and governance risk signals and maps the resulting risk state to seven levels of intervention, ranging from monitoring and output restriction to human review and forced system shutdown. Minimum hold-time rules, authorization boundaries, and auditable records are incorporated to support stable escalation and accountable intervention. The framework is illustrated through three scenarios involving military-drone target misclassification, multimodal prompt injection, and an anonymized medical AI ethics committee. These scenarios illustrate how different types of risk can activate technical controls, human oversight, and governance responses and are not intended to constitute empirical validation of system performance. A structured comparison with IEEE 7009–2024 and ISO/IEC 42001:2023 further illustrates how the TEG framework may complement existing safety and management standards. The study also proposes preliminary response-time targets for future simulation and prototype evaluation. From a policy perspective, TEG provides a configurable architecture that can support resilient AI infrastructure, responsible technological deployment, and institutional accountability, with Taiwan serving as a context for localized governance.