Managed Autonomy at Runtime: Gear-Based Safety and Governance for Single- and Multi-Agent Cyber-Physical Systems
arXiv:2607.00334v1 Announce Type: new Abstract: Autonomous agents, whether LLM-driven software agents or robotic physical agents, face a common class of failure modes when operating without continuous human oversight: safety violations from unverified actions, behavioral instability from unconstrained loops, and continuity loss from unhandled error states. We develop \system{}, a discrete-time control system that combines five execution gears (\Gobs{}, \Gsug{}, \Gplan{}, \Gexec{}, \Gint{}) with utility-gated dispatch and event-driven fallback. For the single-agent case, we prove monotonic stab
The proliferation of autonomous agents (LLM-driven software and robotic) necessitates robust safety and governance mechanisms to prevent failures and ensure reliable operation, driving research into systems like the one presented.
Ensuring the safe and predictable operation of AI agents is critical for their widespread adoption and for preventing unintended consequences in both digital and physical domains.
This research introduces a structured control system for autonomous agents, offering a more reliable framework for managing their behavior and mitigating emergent risks.
- · AI software developers
- · Robotics companies
- · Cyber-physical systems integrators
- · Industries deploying autonomous systems
- · Companies with unverified agent deployments
- · Systems susceptible to behavioral instability
- · Manual oversight roles in autonomous operations
Increased trust and accelerated deployment of single and multi-agent autonomous systems across various sectors.
Reduced incidence of AI-related failures leading to fewer regulatory interventions or public backlash.
The development of industry standards and certifications based on similar robust safety-governance frameworks for AI agents.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI