The Signal
For two years the AI security question was about your people. Who is pasting client data into ChatGPT? The network just flipped the question. Ten member companies spent the last 30 days asking how to govern the AI agents they are building and running themselves.
These are not employees using a chatbot. A payroll processor, a health system, a global manufacturer, and a large insurer all named the same gap. They have agents deployed, or arriving within months, with no clean way to inventory them, scope their access, or limit what they can do.
The buyer moved from consumer to producer. When you build the agent, "block the risky app" stops working. You own the risky app now. Eight of these ten asks landed in the last two weeks. This is where enterprise AI security is heading next: securing machine actors you created, not the humans you manage.
From the Network
"We are currently evaluating how to securely govern enterprise AI agents, control access to internal data, and maintain visibility into how models and agents use sensitive content."
"We are evaluating tooling to govern AI agents, including placing guardrails around agentic actions."
"We have budget for initiatives geared towards implementing AI Security and AI Governance solutions for our FY27 beginning October 1. We are anticipating that this will affect identity, data and SecOps."
Three industries, one job to do. Get their arms around the agents before the agents get access to everything.
Top Open Priorities This Week
Two raw asks pulled directly from member submissions in the last 14 days, unedited:
"Looking for tools to better understand and police what AI agents are being leveraged across the business units."
"We rely on MS Defender, Intune, Entra ID, Skyhigh CASB, and our SecArch and AI Governance processes to identify and govern AI usage. Discovery is primarily control and review based rather than providing automated enterprise-wide inventory of AI agents and MCP."
Both leaders have the governance intent. Neither can see every agent already running in the building.
Member Spotlight: Deb Cafarella, Shields Health
The agent-governance push runs straight into the tension Deb Cafarella lives every day: control the risk without becoming the office that blocks the business. She leads security at Shields Health, and she put the stakes plainly in her DoGood member spotlight, "If you're a department of no and you're not collaborative, people go around you."
The Context
On July 21, Box shipped a set of controls to govern how AI agents touch enterprise content. Guardrails tied to data sensitivity. Permission scoping for outside agents like Claude, ChatGPT, and Gemini. Classification rules that keep agents out of tagged files. In Box's own 2026 enterprise AI survey, 90% of IT leaders said security and trust worries are the top reason they hold agents back from company content.
The headlines are catching up to what the network already knew. Members were asking for agent inventory, scoped access, and guardrails weeks before a major platform shipped them.
Bottom Line: The control layer for agents is forming at the content and data tier, not the network edge. Point your evaluation cycles at the data, because that is where an agent does its damage.
What to Do About It
Pull an inventory of every AI agent running in your environment this quarter, including the ones your own teams built and the ones riding inside SaaS tools. For each, write down what data it can reach and what it can do without a human in the loop. Where an agent has broad access and no approval gate, scope it down before your next audit cycle.
The CXO Brief is powered by the DoGood network, 5,000+ IT leaders sharing what they are actually working on.
Know a CIO who needs this? Forward it and they can subscribe here.
Enterprise IT leader at a $100M+ company? Apply to join DoGood.