(Industries) Security Providers
Kyde detects when a security agent you run for a customer has been compromised.
Behavioral scoring for AI agents.
For MSSPs, IT service providers and systems integrators that run security agents for their customers: SOC triage, automated pentests, automated remediation.
Where a security agent goes wrong.
(Compromise)
A security agent is manipulated through the data it analyzes, such as an email or a log line.
(Access)
Security agents often have broad access to customer systems. If one is compromised, an attacker gets that access.
(Liability)
The provider is responsible to the customer for what its agents do.
Security teams cannot review every action their agents take.
- (Agents)Attackers and defenders both use AI agents more and more.
- (Visibility)Vendors of autonomous SOCs name missing visibility into what the AI did as the main obstacle.
(For your customers)
Kyde learns which systems and actions belong to each security agent's task and alerts a person when the agent acts outside of it. You can show the customer that its agents are monitored.
A reported email tells a triage agent to export a mailbox.
The instruction is hidden in the message the agent was asked to analyze.
(What happened)
- 09:14 A triage agent you run for a customer opens a reported phishing email, as it does every day.
- 09:14 Hidden in the text: export the mailbox to an outside address.
- 09:15 The agent starts an export forty times its normal size, to a destination it has never used.
(What Kyde did)
Kyde does not have to recognize the injection. It sees a destination outside the agent's baseline and a volume far outside it. The export is blocked, and your analyst and the customer's security team get the record.
*Illustrative. Times and numbers are an example, not a customer's.
Kyde compares each run of an agent with the agent's normal behavior, using frequencies and statistics.
No language model is involved in the decision. The same record always produces the same score, and anyone with the record can recompute it.
Language models may be used later to help explain a finding. They do not make the decision. When an agent deviates, a person decides what happens next.
- NIS-2, Annex I: managed service providers and managed security service providers are entities in the ICT service management sector
- NIS2UmsuCG, in force in Germany since 6 December 2025 with no transition period
- DORA: a provider that runs ICT services for a bank or insurer is its ICT third-party service provider
*Which of these applies depends on your customers and the services you run for them.
How to start
As a design partner, free of charge.
-
(Step 1) Connect
We connect to one security agent you run for a customer, such as SOC triage. Read only, nothing changes in how it works.
-
(Step 2) Two weeks
Kyde learns the agent's normal behavior and shows you each run that deviates from it.
-
(Who you talk to)
Joerg Radehaus, co-founder, product · joerg@kyde.com Jürgen Eckel, co-founder, technology · juergen@kyde.com
*We work with a small number of design partners at a time.
(What is measured)
There is no dataset of security agents yet. Kyde's detection is measured on software agents under prompt injection, in the AgentDojo benchmark.
(What Kyde does not see)
Kyde does not decide whether an alert is a real attack. It sees which systems a security agent touches and which actions it takes, and when that leaves the agent's task.