AI-OS: A Meta-Level Operating System Framework for Adaptive Artificial Intelligence
Zvonko Vulicevic · Zenodo (CERN European Organization for Nuclear Research) · 2025
This work introduces AI-OS v1.0, a reference operating-system–level framework for adaptive artificial intelligence systems. Unlike existing approaches that focus on models, algorithms, or safety mechanisms at the application level, AI-OS formalizes system-level regulation as a distinct architectural necessity. The framework defines a meta-level regulatory layer responsible for accessibility, routing, inhibition, execution authority, irreversibility, and degradation under conditions of uncertainty, latency, and partial visibility. Internal generative and learning processes remain unconstrained in content, while authority over execution and irreversible impact is explicitly externalized and regulated. AI-OS is non-prescriptive and implementation-agnostic by design. It does not propose new algorithms, benchmarks, or performance claims. Instead, it provides a stable architectural foundation that prevents the collapse of control into local adaptive logic and enables safe, accountable deployment of long-running AI systems.