Policy-as-code for AI authorization

Decision Function

Decision = f(user, agent, capability, action, resource, purpose, context, risk)

Authorization is evaluated at runtime using all dimensions of the request, not just identity or role.

Policy Evaluation Flow

1. Authenticate user identity
2. Validate user → agent access
3. Validate agent → capability access
4. Apply user data entitlements
5. Evaluate purpose, context, and risk
6. Return allow, deny, or step-up decision

Example Policy (Cedar-style)

permit(
  principal == User::"adam",
  action == Action::"InvokeAgent",
  resource == Agent::"finance-agent"
)
when {
  context.purpose == "reporting" &&
  context.risk != "high"
};