Aegis Cortex (AC-OS): A Zero-Trust Enterprise AI Agent OS

Abstract: When deploying generative Large Language Models (LLMs) in high-risk industrial and core commercial networks, they face a fundamental architectural conflict between their probabilistic generative nature and the extreme deterministic requirements of business environments. To resolve the "architectural weakness" of current Agent frameworks, this project open-sources Aegis Cortex (AC-OS) —a single-loop runtime orchestrator and intelligent agent execution engine based on a zero-trust architecture. AC-OS abandons the traditional evolutionary empowerment paradigm and establishes a physical sovereignty contract of separation of powers: "Cognitive Proposition (Noesis), Governance Arbitration (Aegis), and Underlying Execution (Runtime), " thoroughly stripping the large model of its direct peripheral invocation sovereignty. The system innovatively introduces a fully out-of-band biomimetic governance matrix: it utilizes a Unified Anterior Cingulate Cortex (Unified ACC) to evaluate cognitive logical entropy in real-time, and a Hypothalamus to strictly monitor the second derivative changes of underlying compute token consumption. Combined with a cross-temporal credit scoring engine to calculate the absolute Effective Risk (R₄₅₅₄₂ₓ₈ₕ₄), AC-OS enables physical-level meltdowns and interceptions against slow data exfiltration, advanced persistent jailbreaks (APT-style probing), and logical hallucination deadlocks. Reference Note: The specific risk control policies of the Aegis Ingress/Egress Gates in this system, as well as the underlying Token management architecture based on metabolic and quota concepts, are an evolution of our previous research. For detailed topological definitions and quota allocation logic, please refer to the previously published "Aegis: Enterprise AI Governance Runtime Specification"