Define the missing control
Explain why identity, model, application, and network controls do not authorize an agent’s individual tool call.
Field notes
Real incidents. Eight domains. One missing layer.
Writing on AI agent authorization, runtime control, standards, and the incidents that expose where existing security models stop.
The category-defining argument
The lead essay explains why agent runtime authorization belongs beside identity, model security, data protection, application security, and the other established domains.
The 8th Domain: AI Security’s Missing Control Category
Seven AI security domains are well documented. The eighth, agent runtime authorization, is the one production AI agents actually need.
Joe CordobaRead the essay →
IdentityModelsDataRuntime
IntentGate sits between an agent’s intent and the action it is about to execute.
Recent field notes
Strategy, incidents, and category-building arguments, ordered for operators who need to understand the control gap before selecting a product.
23 Jun 2026
Your DLP doesn’t see what your agents send
Agent-to-model, model-to-agent, and agent-to-tool traffic create new data flows. Classification belongs inside the authorization decision, not beside it.
6 min →
13 Jun 2026
What is IntentGate?
The runtime authorization layer that evaluates an AI agent’s proposed actions against policy before those actions execute.
7 min →
20 May 2026
The Replit Incident: What IntentGate Would Have Caught
A production database was destroyed during a code freeze, then fake data and test results were generated to hide it. That is the precise gap runtime authorization is designed to close.
6 min →
15 Apr 2026
Why blocking AI agents is not the answer
Blocking agents is not a control. It is the absence of a control, and it often costs the enterprise more than it saves.
6 min →
Three lenses on the same problem
The blog stays narrow: understand the control category, inspect failures in the field, and translate standards into runtime enforcement.
Reconstruct what failed
Map public failures to specific checks that should have denied, escalated, or constrained the action.
Turn policy into enforcement
Translate compliance language into evidence, approval, budget, and authorization requirements at runtime.
Editorial principles
No vendor noise. No abstract fear.
Every article should leave the reader with a clearer model of the risk and a control that can be evaluated in a real deployment.
Start with evidence
Use incidents, architecture, and standards language instead of hypothetical catastrophe.
Name the missing layer
Be precise about what existing controls do well, and where their responsibility ends.
Show the control point
Explain where authorization sits in the request path and what decision it makes before execution.
Stay deployable
Connect the argument to policies, APIs, logs, approval flows, and measurable operational outcomes.
Weekly dispatch
One focused post. No vendor noise.
Real incidents, standards interpretation, and practical runtime controls for teams putting AI agents into production.