AIOA Whitehat: Epistemic Control System

AI risk increases when AI systems move from giving advice to taking actions: writing code, changing files, calling tools, using providers, browsing the web, installing packages, or operating Git without clear human control. AOIA-Core reduces this risk by creating an epistemic control layer between AI suggestions and real-world action.The system separates suggestion from authority. A model can propose, but it cannot approve, execute, write, push, call providers, or satisfy a gate by itself. Every meaningful action must pass through review, preview, evidence checks, hash-bound records, and an explicit human approval barrier. This makes hidden or confusing AI behavior easier to inspect before anything irreversible happens.The impact is both practical and research-oriented. In the near term, AOIA-Core can help developers and small teams use AI coding agents more safely by preventing uncontrolled execution, unsafe Git operations, package installs, provider calls, and metadata-as-authority mistakes. In the longer term, the same pattern can become part of a broader safety architecture for human-supervised AI agents.This does not solve AI x-risk alone, but it targets an important failure mode: loss of human control over increasingly capable AI workflows. By making AI actions inspectable, auditable, and explicitly approved before execution, the project aims to reduce the chance that AI systems silently escalate from advice into autonomous control.