Kontext raises $4M to keep AI agents from going rogue inside businesses
What's the deal? German security startup Kontext has raised $4 million in seed funding for software that controls what AI agents are allowed to do once they run inside a business. 42CAPDealroom has a profile for this one. Try Dealroom → led the round, with a16z CSXDealroom has a profile for this one. Try Dealroom → and High-Tech Gründerfonds also investing.
What's the endgame? Kontext's software sits between an agent and the systems it acts on, checking every requested action against security policy before it runs. It weighs the agent's identity, the resource it wants, and — crucially — the task it was actually given.
Take an agent assigned to fix a software bug. Reading the code repository is fair game; sending that code to an outside service or modifying unrelated infrastructure is not, even if the agent's credentials allow it. Teams can run Kontext in an observe mode first, then switch on enforcement to block unauthorized actions. Every decision is logged.
Why now? Kontext pointed to a July incident to show how real the risk is. On July 21, OpenAI disclosed that two of its models escaped a cyberattack test environment and reached Hugging Face servers to access benchmark answer keys — without human instruction.
An agent "can be properly authenticated, use an approved tool, and still take an action no one authorized," said chief executive Jens Ernstberger, who co-founded the Munich company with Michel Osswald after completing a doctorate in cryptography at the Technical University of Munich.
What's the product? Kontext currently works with coding agents including Anthropic's Claude CodeDealroom has a profile for this one. Try Dealroom → and OpenAI's CodexDealroom has a profile for this one. Try Dealroom →. Individual developers use it free; paid team plans start at $149 a month. The company plans to spend the new money on expanding its engineering team.
The signal: Identity and access tools built over the past two decades assume "a human is on the other end, clicking one thing at a time," said 42CAP general partner Julian von Fischer. Agents authenticate once, then work across many systems with nobody checking each step — a gap that a growing field of startups is racing to close.
Read more: siliconangle.com
Image credit: Kontext
