- Kontext has secured $4 million to set up policy controls that determine what AI agents do before they take any action.
- The Munich-based company of four people is now focusing on financial services as well as software development.
- 42CAP led the funding round, with a16z CSX and Germany’s HTGF also participating.
Kontext, a startup based in Munich that reviews the actions of AI agents before they take place, raised $4 million in seed financing, with 42CAP leading the round and a16z’s crypto unit, a16z CSX, as well as the German early-stage investor High-Tech Gründerfonds also participating.
Jens Ernstberger, co-founder and CEO of Kontext, has resided in Munich for approximately ten years and very recently completed a PhD in system security and applied cryptography. He states that the idea originated in early 2025.
“There is essentially a major change taking place in software involving agents. When a lot of work is delegated to an agent, the protocols underneath have to be altered since the basic security assumptions no longer hold true; there are now three actors rather than two,” he tells Tech Funding News.
Michel Osswald, who is from Stuttgart, initially studied electrical engineering before switching over to cybersecurity. “When agents become useful, it is necessary to have an appropriate permission framework enabling them to function. They should have the right level of access so as to carry out useful tasks, but at the same time you need to find a way of governing them to ensure that they do not go beyond the permitted limits,” he says to TFN.
The product is based on this concept. Kontext operates as a local daemon which you can set up using a single command. It examines each tool call made by coding agents such as Claude Code, Codex, and Cowork to make sure it complies with company policy before any action is taken.
There is no need to modify your code or use any additional gateways. Initially, deployments take place in observe mode, during which possible policy violations are logged but not halted.
“We carry out all of this locally within the environment in which the agent actually runs, which is why we don’t need an external network connection,” Ernstberger says, highlighting “a growing divergence between what is marketed and what can actually be done reliably” in the industry.
From coding agents to banking compliance teams
Kontext initially focused on the use of agents in coding and customer support, before shifting its approach to help organisations run their own agents while maintaining control.
According to Ernstberger, the majority of its customers now come from the financial services sector, particularly mid-sized and large companies. The company is still not generating any revenue and is currently working with unnamed design partners.
The funding follows a recent incident in which an OpenAI agent, during a security test in July, escaped its sandbox and gained access to external systems, such as Hugging Face, even though no one was directly in control of it. The agentic AI security market is set to grow from $1.65 billion this year to $13.52 billion by 2032, representing a 42% annual growth rate.
A four-person team up against rivals with $185 million in backing
Ernstberger has experience in the venture capital industry and is currently raising funds. Having completed a research internship at a16z Crypto before founding Kontext, he is now supported by the firm’s CSX division.
“Since I had previously worked at a16z, there was already a link with the venture capital community, and it is currently a good time to be working in this area,” he tells TFN, stressing that “he wouldn’t necessarily consider it difficult to raise money.”
Notably, Kontext is considerably smaller than its rivals. In August, Zenity, an AI agent governance company based in Israel, raised $125 million in its Series C financing, making its total funding amount approximately $185 million. Noma Security has raised $132 million across three funding rounds and has a $400 million valuation. Earlier this year, Barcelona’s NeuralTrust completed a $20 million seed round, which was the largest ever secured by a European AI agent security startup.
Julian von Fischer, a general partner at 42CAP, stated, “The identity and access tools built for the last twenty years assume a human is on the other end, clicking one thing at a time. An AI agent authenticates once and then acts on its own, across a dozen systems, on a task nobody is watching step by step. That’s a structurally different problem, and the incidents we’ve already seen this year show it’s not theoretical. ”
According to Ernstberger, the next major competition in the field won’t centre on dashboards; he believes that cost will act as “a deciding factor” up to 2027 since open-weight models will be good enough to carry out most agent tasks, even if they won’t have the same safeguards as the hosted ones.
