A policy framework for deploying autonomous agents in public-safety, defense, and critical-infrastructure settings. Co-authored with three federal advisors.
Agentic AI in regulated environments forces a different conversation than enterprise productivity AI. The question isn't "is the agent helpful?" It's "can we account for every action the agent took, and override it when we need to?"
This paper proposes a policy framework — Innovent's deployed-customer pattern — covering five elements: typed policy enforcement, attended-vs-supervised modes, decision registers, override authority, and continuous audit.
The paper is written for the people who have to sign off on autonomous decision-making in their domain — CISOs, compliance directors, civilian oversight bodies, defense acquisition teams.
It's not a technical specification. It's a framework for talking about agentic AI in language that risk and policy teams already use.
Includes deployment topology diagrams, reference architectures, and the full 30-item self-assessment.
Request the PDF