Architecture-Level Safety Enforcement for Autonomous Systems through Non-Bypassable Execution Control

This paper introduces an architecture-level framework for safety enforcement in distributed autonomous systems. It positions safety enforcement as a core architectural layer within a broader governance architecture, complementing capability governance and execution control mechanisms. Related works: - Safety-Bounded Autonomy: Architectural Safety Enforcement for Distributed Autonomous Systems - A Governance Architecture for Safe and Bounded Autonomous Systems - Non-Bypassable Execution Control in Autonomous Systems - Capability-Gated Autonomy in Distributed Autonomous Systems Autonomous systems are increasingly deployed in complex and safety-critical environments. As these systems scale in autonomy and operate in distributed multi-agent configurations, ensuring enforceable and verifiable safety constraints becomes a central architectural challenge. This paper establishes an architecture-level approach to safety enforcement in distributed autonomous systems. The proposed framework defines safety enforcement as a system-level property, in which all executable actions are constrained by enforceable safety boundaries embedded directly within the system architecture. Rather than relying solely on behavioral correctness, policy validation, or post-hoc monitoring, the framework introduces capability-gating as an architectural enforcement principle. Candidate actions generated by decision-making components must satisfy predefined safety contracts and system constraints before execution, ensuring that only admissible actions can be executed within the system. By separating decision generation from execution admissibility and runtime safety enforcement, the architecture ensures that distributed agents can operate autonomously while remaining within verifiable safety limits. The framework is designed to support heterogeneous robotic platforms and scalable multi-agent deployments. It enables the integration of both local safety constraints and shared system-level safety contracts, allowing coordinated safety enforcement across distributed systems and reducing the risk of unsafe emergent behavior. This approach establishes a structural foundation for safety-bounded autonomy, in which safety is enforced through architectural constraints rather than assumed through correct decision-making alone. Safety enforcement is thus treated as an intrinsic system property rather than an external validation layer. This work contributes to the emerging field of architectural safety enforcement in autonomous systems and provides a foundation for scalable, verifiable, and enforceable system behavior in real-world deployments. This paper focuses on the safety enforcement layer within a broader architectural stack comprising capability governance, execution control, and runtime safety enforcement. This work is part of a broader research program on governance and execution control in autonomous systems, including capability governance, safety-bounded autonomy, and non-bypassable execution control architectures.