How a Leading Global Private Equity Firm Went From Blocking AI Agents to a Secure Enterprise Program in Days, Not Months

A leading global private equity firm with approximately $150B in assets under management was facing the kind of pressure that travels quickly in private equity: identify the operational advantage in a new technology early, without exposing the business to unnecessary risk. AI agents had become part of that conversation across the firm, but none had yet been approved for use in production.

With strict regulatory obligations, client data privacy requirements and complex compliance mandates, the firm needed a way to say yes to agents without flying blind.

The security engineer responsible for making that happen has over a decade of infrastructure and security experience, and manages AI agent security along with multiple other security programs.

Geordie was deployed across the entire estate at the firm, on every laptop and endpoint in the organization, with no disruption to end users and no changes to existing infrastructure. The scheduled task architecture meant there was no persistent agent running on machines, and the absence of SSL inspection meant the existing network security stack continued to operate as designed.

The first meaningful finding came almost immediately. During the initial deployment, Geordie surfaced the extent to which developers were using alternative agent providers and sub-models inside their development environments, which is activity the security team had no prior visibility into.

The implementation also established a new AI agent approval workflow. Rather than reviewing individual agents, which would quickly become unmanageable at scale, the team built a platform-level approval process. For each new agentic platform under consideration, the security engineer would review the activity with Geordie, including the prompts, risk flags and connection data, then produce a documented assessment for the CISO and the firm's AI working group.

I go through the platform in Geordie, look at the chats, the prompts, any risk indicators that surface, and I give a clear assessment to my team. Everyone can see that someone has looked at it properly, that we have visibility, and that we're accepting informed risk rather than flying blind.

As the firm's AI agent program has matured and the volume of agent activity across the environment has grown, one conviction has become central to how the security team thinks about risk: the only view that actually tells you what is happening is the view from the agent itself.

The security engineer is clear that this is not a theoretical concern. In practice, the most common source of agent risk is not malicious intent, it is well-intentioned users who are comfortable with the end result and unaware of what the agent did to produce it.

Someone builds a perfectly reasonable workflow, runs it every day, checks that the output looks right, and has no idea that their agent has been quietly doing something they never intended. You can't see that from the network layer. You can't see it from the identity layer. You can only see it if you're watching the agent, from the Geordie perspective.

This is the lens through which the firm evaluates every new agent deployment and every new security requirement. Not whether a tool can be blocked at the boundary, but whether the security team can see what the agent is doing once it is inside in terms of how it reasons, what it connects to, how context flows across the workflow and when behavior moves outside what was originally intended.

For a firm operating in financial services, where client data, investment information and regulatory obligations are at stake in every workflow, that perspective is not optional. It is the foundation on which every approval, every compliance assessment and every executive conversation is built.

My leadership team approved these tools because I could show them with Geordie what the agents were doing. Not what the model was doing. Not what the network was logging. What the agent was actually doing, step by step, on their behalf. That's the view that matters.

Ask the security engineer what has made the Geordie relationship work and the answer is consistent: the team.

In fifteen years of enterprise security work, he had never described a vendor relationship as a genuine partnership. Most of the time, feedback goes in and nothing comes back. Feature requests are acknowledged and forgotten. Account teams cycle through and institutional knowledge disappears. The relationship with Geordie has been different from the start, and that difference has had a direct impact on what the firm has been able to accomplish.