AI for insurers, behind the risk perimeter.
Claims, policies and underwriting are document work. ARK puts open vision and language models on that work inside your own infrastructure, so policyholder data stays where your risk framework says it must.
of context on the largest Qwen models: a whole underwriting submission in one request, without chunking.
Read moreto the first token of a new 64,000-token file. Ask it a second question and the answer comes 9× sooner.
Read moreWith a second copy of each part of the model, a failed server doesn't take the service down. Requests in flight are replayed on another path, and servers join or leave without a restart.
Read moreWhat stands between insurers and production AI.
In a 2026 survey skewing to Fortune 500 and Global 2000 companies, 92% ranked data privacy, explainability and compliance as top vendor-selection factors, ahead of performance at 74%.
Plug and Play 2026 Enterprise AI Strategy Pulse Survey, via Forbes, 6 August 2026.
Documents everywhere
Claim forms, damage photos, invoices, medical reports and policy wordings arrive in every format.
Sensitive personal data
Policyholders' health and financial details are among the most protected data under GDPR. They can't be sent to an outside model provider.
Pricing models become high-risk AI
The EU AI Act classes AI used for risk assessment and pricing in life and health insurance as high-risk. Those obligations apply from 2 December 2027.
EU AI Act, Annex III point 5(c); dates under Regulation (EU) 2026/1744.
Costs you can't forecast
Claims and underwriting workflows make several model calls per case. One insurtech CEO warned that a provider's price change could double or triple the cost of AI work.
Insurance Business, 16 October 2026.
What insurers build on ARK.
Each of these runs on open-weight models inside your infrastructure, or on ARK Cloud in the EU while you prototype with synthetic data.
Claims intake
Read claim forms, invoices and damage photos with the Qwen vision models and prefill your claims system with structured JSON.
Policy questions
Answer agents' and customers' questions grounded in your own policy wordings, with retrieval over bge-m3 embeddings.
Underwriting files
Summarise long submissions in a single request with a 256K-token context. A 64,000-token file reaches its first token in 4.7 seconds.
Correspondence
Draft letters and emails in the customer's language for a handler to review.
What your risk and compliance teams will see.
Inside your network
With ARK Core, the gateway, compute servers and models all run on your infrastructure, on NVIDIA GPUs you choose. Your team operates them.
Logs without content
Every request carries a correlation id and reports its token usage. Prompt fields are redacted by default, and generated text is not logged.
Known, open models
Every model is open-weight, and its licence is published. The model that answered is named in every response.
A real exit plan
Open-weight models behind OpenAI- and Anthropic-compatible APIs mean the same models and code run elsewhere if you ever need to leave.
Built to stay up
ARK splits each model into stretches of layers spread across your GPUs and machines, and can hold a second copy of any stretch. If a server fails mid-request, the request is replayed on another path and the service keeps running. Servers can join or leave a running deployment without a restart.
The rules an insurer's AI works under.
ARK keeps processing inside the infrastructure you control and gives your governance the facts it needs. Your compliance assessment stays yours.
Under the EU AI Act, AI used for risk assessment and pricing of natural persons in life and health insurance is high-risk (Annex III), from 2 December 2027. For AI that is not high-risk, EIOPA's August 2025 Opinion expects proportionate, risk-based governance under Solvency II and IDD.
Two ways to run ARK. One API.
ARK Core For insurers
The platform and its models run on your NVIDIA GPUs, licensed per GPU and operated by your team. Our engineers can size it with you, prove it on your hardware and support the rollout.
ARK CoreOur EU infrastructureARK Cloud
Prototype with synthetic data on the hosted API, then move in-house without code changes.
ARK CloudStart with one claims or underwriting workflow.
Bring a sample of claim forms or a submission file. We run it on ARK, on your GPUs or in our EU cloud, and show you the results and the numbers before you commit.