Research and governance framework for defining what AI agents can do, when humans must review, and what evidence is required.
The Agent Authority Framework is a research and governance system for autonomous systems in high-trust environments. It defines the operating rules for AI agents — what actions are permitted, what context is trusted, when human review is required, and what evidence must be captured. The framework is being developed as both a conceptual model and an implementable system through AgentPlugin.
AI agents are moving from content generation to execution — they can send messages, update records, make decisions, and trigger workflows. But most organizations have not defined the operating rules: What is the agent allowed to do without approval? Which data can it trust? When must a human review? What evidence must be captured? Capability without authority creates operational risk.
The Agent Authority Framework provides a structured model for answering these questions. It is organized around seven gates that every agent action must pass through: allowed actions, system access, business context, risk assessment, human review requirements, evidence capture, and ownership. Each gate defines specific criteria that must be satisfied before an agent is authorized to act. The framework is designed to be implemented in software through AgentPlugin.
Developing the theoretical framework and governance model.
Writing and publishing the framework for public and client use.
Core framework documented. Seven Gates model published. Integration with AgentPlugin in progress. Industry-specific applications being researched.
AI agent tooling and plugin architecture — structured context, permissions, and evidence capture for autonomous systems.
In DevelopmentRelationship and meeting intelligence — structured context routing, qualification, and asynchronous discovery for high-stakes conversations.