(Industries) Insurance
Kyde detects when a claims or service agent changes how it decides.
Behavioral scoring for AI agents.
Where an agent in claims or service goes wrong.
(Compromise)
A claims agent is manipulated through a document it reads.
(Drift)
After a model update, a claims agent approves cases it used to escalate.
(Data)
A service agent passes on customer data that its task does not require.
Insurers are starting to use agents in claims and service.
- (KI-MIG)Since 29 July 2026, BaFin supervises AI used in regulated financial activity, insurers included.
- (DORA)DORA has applied to insurers since 17 January 2025. In December 2025, BaFin made clear that AI falls under its ICT risk management, with testing and the management of third-party providers.
(For underwriters)
The same score shows how a customer's agents behave over time. Insurers who cover AI agents can use it when they price the risk.
AI agent insurance, in depthA claims agent approves every claim after a model update.
It keeps running, completes its tasks and reports success.
(What happened)
- 23:00 A claims agent starts its nightly batch after a model update.
- 23:20 Normally it approves six claims in ten. Tonight it approves ten in ten.
- 02:00 Left alone, it would have paid 1,318 claims by morning.
(What Kyde did)
The approval rate leaves its band after the first forty claims. Kyde puts the agent in quarantine and hands the open claims to a person.
*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.
- DORA, ICT risk management and third-party risk, for insurers since 17 January 2025
- Solvency II, system of governance and risk management
- KI-MIG, the German AI market surveillance act in force since 29 July 2026: BaFin supervises AI used in regulated financial activity, insurers included
- GDPR Article 22, decisions based solely on automated processing
- EU AI Act: motor claims are not Annex III high-risk. Point 5(c) covers risk assessment and pricing in life and health insurance. The Article 50 transparency duties still apply
*Which of these applies depends on the process the agent runs. Kyde is built for them. It does not make a process compliant.
How to start
As a design partner, free of charge.
-
(Step 1) Connect
We connect to one agent you run in claims or service. 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)
Kyde's detection is measured on software agents under prompt injection, in the AgentDojo benchmark. We have no measurement on a claims agent in production yet. That is what we do with design partners.
(What Kyde does not see)
Kyde sees what an agent does: its tool calls, the data it reads, where it sends it and how often. It does not judge whether a single claim was decided correctly. It sees when the agent's decisions stop following their usual pattern.