Agentic AI , Artificial Intelligence & Machine Learning , Next-Generation Technologies & Secure Development

The winner of this year's RSAC Innovation Sandbox contest raised $30 million to provide visibility into artificial intelligence agents across cloud, code and endpoint environments.

See Also: AI Agents Introduce a New Insider Threat Model

The Balderton Capital-led Series A round will provide London-based Geordie AI with the resources needed to expand engineering capabilities, accelerate product development and maintain the pace of innovation customers require, said co-founder and chief AI officer Hanah-Marie Darley. She said Geordie AI is developing a security framework designed specifically around AI agent behavior and governance.

"Agents are inherently operating across all of these surface areas, often simultaneously," Darley told ISMG. "When you're talking about security, you can't just have something on the endpoint, you can't just have something in cloud solutions. You need something that has coverage across cloud, code and endpoint services, because that's where agents are operating day to day."

Geordie AI, founded in 2025, employs 37 people and has raised $36.5 million, having previously received $6.5 million in a September 2025 seed funding round co-led by Ten Eleven Ventures and General Catalyst. The company has been led since its inception by Henry Comfort, who previously spent five-and-a-half years at Darktrace, culminating in a role overseeing global business operations (see: OpenAI's Daybreak Bets on Agentic Cyber Defense).

Why Organizations Struggle With AI Agent Visibility

Agents are becoming part of core business infrastructure, Darley said, automating workflows, making decisions and executing tasks that were previously handled by employees. Organizations, therefore, can't rely on existing controls designed for users, applications and infrastructure operating within established perimeters. AI agents behave differently, operate autonomously and continuously evolve.

"This is not just another small control plane," Darley said. "It's not just a small layer or a security tool. It's really changing business infrastructure. It's really changing materially how businesses operate. And so the type of security paradigm that you need is not going to be a simple tool layer or something that is just going to sit on a point solution that's going to be off to the side."

Organizations often underestimate how different AI agents can be from one another, with a Claude-based agent, a Copilot Studio agent and a custom-built enterprise agent all operating differently, using different tools and creating different security risks. And new agent frameworks appear regularly, forcing security teams and vendors to continuously adapt their visibility, monitoring and governance strategies.

"That surface area is going to continue to expand, and we need to be right at the front of that," Darley said. "It involves consistent coverage and security as we evolve not only our models and the way that we continue to prioritize proprietary risk and enforcement, but also the way that we continue to stay ahead as the space changes."

Many organizations currently lack even a basic understanding of what their AI agents are doing on a day-to-day basis, with existing tools not capturing tool usage, agent-to-agent interactions and autonomous workflows, Darley said. Geordie AI's approach involves profiling agents, establishing behavioral baselines and performing anomaly detection, which helps organizations more easily identify unusual behavior.

"One of the things that we continuously do is essentially profile agents," Darley said. "We understand how they operate. We give you behavioral baselines, and we also go into anomaly detection. And that's only going to evolve as we continue to deepen not only our understanding and our customers' understanding of agents, but also as the space continues to evolve and change."

How Routing Activity Through Gateways Introduces Complexity

Security architectures that rely heavily on gateways and proxies can provide value during early-stage deployments, but they become increasingly difficult to manage as organizations expand their use of AI agents, Darley said. As agents access more tools, APIs and SaaS services, routing all activity through centralized gateways can introduce operational complexity and performance bottlenecks, she said.

"Instead of having to architect all of your workflows through a gateway, you're able to operate with policy enforcement across all of these agents," Darley said.

Just as human employees need policies, procedures and training to operate safely, AI agents require contextual information that guides their decisions in machine-readable formats and often need it delivered in real time, Darley said. Security and governance controls must therefore be integrated directly into agent workflows rather than delivered through periodic training or static documentation, Darley said.

"For us, that also centers on context engineering," Darley said. "Giving the agents back the same context that you would give your security engineers, but in an agent-readable way."

The industry is beginning to move beyond single-agent deployments toward more dynamic models in which agents autonomously hand off tasks to one another and coordinate activities across multiple systems, Darley said. Organizations must therefore understand not only individual agent behavior but also how agents interact with each other, exchange information and collectively execute workflows.

"You have to have purpose-built security when it comes to agents," Darley said. "You won't be able to just pivot something from your existing solution, because you'll have gaps - whether that's context gaps, whether that's security control gaps, or whether that is a combination of visibility and control gaps. The key challenge around agents is that they're not the same as traditional software."