Kontext raises $4M for runtime security platform for AI agents
Tech.eu Tamara Djurickovic
Kontext just raised $4M to lock down AI agents while they work. Its pitch: stop a “trusted” bot from doing something nobody actually approved.
Based on reporting by Tech.eu, Tamara Djurickovic — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Kontext has raised $4 million to push its runtime security platform for AI agents beyond the lab and into real workplaces. The round was led by 42CAP, with backing from a16z CSX and HTGF.
The company was founded by Jens Ernstberger and Michel Osswald, who bring backgrounds in secure computing, applied cryptography and AI systems. Their focus is simple enough to say and messy enough to build: as AI agents move from chat windows into company software, they start handling code, files and credentials, and that’s where the security problem gets real.
Kontext sits between the agent and the tools it wants to use. It checks activity in real time against policy and risk signals, looking at things like the agent’s identity, the requested action, the target resource and the task it was assigned. The twist is that an agent can be authenticated, use an approved tool and still try to do something nobody asked for.
That’s why the product is built around the moment of action, not just login and access. Teams can run it in observe mode first, which lets them watch how agents behave and see where policies would bite without stopping work. Once enforcement is turned on, the platform can block unauthorized actions before they happen and keep an audit trail of each decision.
The new money will go into engineering, continued development of the runtime enforcement platform and customer support for teams deploying agents with more visibility and control. It’s a pretty direct bet that the next AI security fight won’t be about model output quality. It’ll be about what the machine was allowed to do in the first place.
My take — AI-written commentary, not fact-checked reporting
This is the right problem to chase, because “AI agent security” without runtime enforcement is mostly vibes and hope. The industry keeps handing agents more keys and then acting surprised when keys are used. Kontext sounds like a sober answer to a very unserious deployment habit.
Read more about this at: Tech.eu