Kontext Security launches with $4m for AI agent runtime controls

A new security company, Kontext Security, came out of stealth with $4 million in funding to sell runtime controls for AI agents, according to a report published by SecurityWeek on 24 September 2026.

SecurityWeek described the company's product as a runtime enforcement platform that evaluates AI agents in real time, and said it is intended to give customers visibility into and control over the actions those agents take. Runtime enforcement, in this context, means checking what a piece of software is doing while it is doing it, rather than only reviewing its configuration or code beforehand.

That framing puts Kontext in the part of the market that treats an AI agent less like a chatbot and more like an unpredictable user account. Agents built on large language models can call tools, browse, write files and trigger other systems, so the question of what an agent is permitted to do at the moment it tries to do it has become a distinct product category. The pitch, as reported, is that organisations deploying agents currently lack a reliable view of those actions.

The SecurityWeek report gives the funding figure as $4 million. It does not, in the material available, set out the investors, the company's headquarters, its headcount, its pricing, or whether the platform is generally available or still in early access. Nor does it describe how the enforcement works in practice: whether controls sit in the agent framework, at the network layer, in front of tool and API calls, or in some combination. Buyers evaluating this class of product will want those details, because the placement of the control point determines what can actually be blocked.

The amount raised is small by the standards of enterprise security funding, and is consistent with an early-stage round rather than a growth round. It is enough to build and sell a first product to design partners, not to fund a broad go-to-market effort.

The Agentic Times has not independently verified the company's claims about its platform, and the reported description comes from the company by way of SecurityWeek rather than from customer deployments or third-party testing. As with other entrants in agent security, the practical test will be whether the controls hold up against agents that are prompted, by attackers or by accident, into doing things their operators never intended, and whether enforcement can be applied without breaking the workflows the agents were bought to automate. No benchmark results, customer names or independent assessments were included in the report.