Geordie AI launches Beam to rein in risky AI agents

Geordie AI has launched Beam, a product designed to manage risk in AI agents by keeping them within defined policies as they make decisions.

Beam is a remediation suite that uses what Geordie calls context engineering to analyse agent behaviour and feed controls back into the system in real time. The London-based company says it is built for AI agents that act autonomously across tools and systems, rather than for conventional software or human users.

The launch comes as security teams grapple with the spread of AI agents inside large organisations. Geordie cited research showing broad use of coding agents among developers, along with Microsoft findings that more than 80% of Fortune 500 companies are using agents in some form.

That adoption has heightened security concerns. Gartner data cited by the company shows that 74% of security leaders see AI agents as a new attack vector, underscoring fears that existing controls may not suit software that reasons, takes action and carries context between systems.

How it works

According to Geordie, Beam starts by mapping how an agent is configured and how it behaves in real time. This includes what the agent is trying to do, what it has already done and the environments in which it is operating.

It then applies policy and contextual controls on an ongoing basis. The goal is to detect risky behaviour and correct it while the agent is operating, rather than relying only on static rules, proxies or gateways placed in front of the system.

Geordie argues that proxy-based approaches can introduce delays and create visibility gaps when agents move between tools. Beam is positioned instead as a way to govern behaviour continuously as agents interact with different systems and make decisions.

The product is intended to support broader deployment of agents beyond trial projects. As more agents run, the system is meant to build a clearer picture of behaviour over time, which Geordie says can improve governance and oversight.

Market traction

Alongside the launch, Geordie disclosed recent growth figures. It said the number of secured agents on its platform increased tenfold in just under five months, while revenue rose tenfold over the past two months.

It also said it had been named a representative vendor in Gartner's Market Guide for Guardian Agents and had won the Black Hat Innovation Spotlight competition in London.

Geordie is part of a growing group of cyber security firms adapting established security concepts to AI systems that can initiate actions on their own. The market has expanded as companies move from experimenting with chatbots and copilots to deploying more autonomous agent-based tools for coding, operations and internal workflows.

Unlike conventional applications, agents can chain tasks together, call external tools and retain context from prior actions. Those traits can increase efficiency, but they also raise the risk of agents exceeding permissions, mishandling data or taking unintended actions if guardrails are weak.

Geordie says Beam is designed to address that issue by shaping agent behaviour during operation. It presents the product as a way to balance security controls with the need to avoid slowing AI systems through extra infrastructure layers.

"Our customers want to actively address the security issues in their AI agents so their organizations can innovate quicker," said Henry Comfort, CEO and Co-Founder, Geordie AI. "Geordie helps them do this with a context-driven 'easy button' that takes care of risks before they snowball into massive problems."