This year’s recipient of the RSAC Innovation Sandbox contest has successfully secured $30 million in funding, aimed at enhancing visibility into artificial intelligence (AI) agents across various environments, including cloud, code, and endpoint ecosystems. This significant investment is poised to bolster the operational capabilities of London-based Geordie AI.

Led by Balderton Capital, the Series A funding round reflects confidence in the startup’s vision for a comprehensive security framework that caters specifically to the behavior and governance of AI agents. Hanah-Marie Darley, co-founder and chief AI officer, elaborated on how the influx of capital will enable the company to expand its engineering resources, accelerate product development, and keep pace with the innovation demands of its clientele. “Agents are inherently operating across all of these surface areas, often simultaneously,” she emphasized, pointing out the necessity for a security solution that spans cloud, code, and endpoint services.

Founded just a year ago in 2025, Geordie AI has already secured a total of $36.5 million, which encompasses an earlier seed funding round of $6.5 million co-led by Ten Eleven Ventures and General Catalyst. Henry Comfort, a veteran with significant past experience at Darktrace, has been steering the company since its inception.

The Challenges of AI Agent Visibility

Darley explained that organizations are increasingly incorporating AI agents into their core business operations, automating workflows, making decisions, and performing tasks once managed by human employees. As a result, existing security measures designed for user and application control within well-defined perimeters are proving inadequate. AI agents not only function autonomously but also adapt continuously, posing unique challenges to security.

“This is not just another small control plane,” Darley stated. “It’s not just a layer or a security tool. It’s fundamentally transforming how business operations are conducted.” Consequently, the traditional security paradigms must evolve beyond simple tools or point solutions to effectively address the complexities introduced by AI agents.

Organizations often misinterpret the variety among AI agents; for instance, a Claude-based agent, a Copilot Studio agent, and a proprietary enterprise agent each operate differently and pose distinct security risks. The constant emergence of new agent frameworks necessitates that security teams and vendors adapt their visibility, monitoring, and governance strategies. “That surface area is going to continue to expand,” Darley warned, emphasizing the need for organizations to develop comprehensive and consistent security protocols to keep pace with this evolution.

Currently, many organizations lack fundamental insights into their AI agents’ day-to-day activities. Existing monitoring tools often fall short of capturing critical metrics such as tool usage, agent interactions, and autonomous workflows. Geordie AI is addressing this gap by profiling agents, establishing behavioral baselines, and applying anomaly detection techniques to help organizations identify unusual behavior. Darley highlighted, “We continuously profile agents to understand their operations, providing behavioral baselines that are pivotal in detecting anomalies.”

The Complexity of Routing Activity through Gateways

The reliance on gateways and proxies in security architectures can complicate the management of AI agent activities as organizations scale their use of these agents. Darley pointed out that while such architectures may provide initial value in early-stage deployments, they can lead to operational complexities and performance bottlenecks when agents begin accessing a broader array of tools, APIs, and SaaS services.

To avoid the pitfalls of routing all activity through centralized gateways, organizations should implement policy enforcement that accommodates the broad usage of AI agents. Much like human employees, AI agents require clear policies and contextual information guiding their decisions, ideally delivered in real-time. Darley stressed that security controls must be embedded into agent workflows, rather than being provided through static documentation or periodic training sessions.

“For us, this revolves around context engineering,” she noted, underscoring the importance of delivering security context similar to what security engineers receive, but in a format understandable by AI agents.

The industry is also shifting focus from single-agent deployments towards more sophisticated frameworks where agents collaborate autonomously to complete tasks by handing off responsibilities between themselves. This necessitates a comprehensive understanding not only of individual agent behaviors but also of their interactions, information exchanges, and collective workflows. “Purpose-built security is imperative when dealing with agents,” Darley asserted, cautioning against the assumption that existing solutions can be repurposed without experiencing significant gaps in security and visibility.