Blog/Private AI for insurance claims processing without sending claims to a third party
Private AI

Private AI for insurance claims processing without sending claims to a third party

By Samuel Seidel · September 9, 2026 · 8 min read

An LLM on a dedicated machine can summarize claim intake, check a claim against policy terms and draft the first response without sending claim files to a third-party AI API. An adjuster's inbox holds medical records, repair estimates, police reports, bank details and the claimant's account, all tied to a name and policy number. The claimant never agreed to an AI vendor processing any of it.

Not legal advice. Insurance is a heavily regulated industry and claims data usually includes categories subject to specific protection rules depending on jurisdiction (health data especially). This post describes an infrastructure choice, not a compliance opinion. Check with your own counsel and compliance function before using AI anywhere in a claims workflow.

What makes a claim file a different category

A claim file combines several sensitive categories at once: health information for a medical or injury claim, financial account details for payout, and a first-person account that can include details about the claimant's home, family, or habits that go well beyond what's needed to process the claim itself. Health data is a special category under GDPR and under sector rules in most regulated insurance markets, which raises the bar for how it can be processed, by whom and on what legal basis.

Third-party AI products aren't built around that bar. A general-purpose API has one data processing agreement for every customer's use case, not one written for claim files with health data. The claimant, who never agreed to any AI platform's terms, has even less visibility into where the file ends up.

What running the model yourself changes

On a dedicated DGX Spark, claim documents go to your own model on a machine assigned only to you, not through a shared third-party inference endpoint, and you set retention on that instance to match your claims system's schedule. It's the setup the private LLM hosting page describes for regulated work, applied to claim files. A rented Spark is still a GPUwerk-run machine (its own in Czechia, or a partner's in the country the console shows), so GPUwerk is a processor under a DPA. Health data in claim files needs your compliance function's approval before any real file goes on the machine.

It doesn't change who may see a given claim. Wire your claims platform's access controls into whatever sits in front of the model, so the AI layer doesn't become a way around role-based access that took real work to set up.

A concrete example

An auto claim comes in with a police report, two repair estimates, and photos with a written description. The model summarizes the intake into a structured form: date, parties, estimated damage range, whether the documents are complete against the policy's required list, and a plain-language summary the adjuster reads in thirty seconds instead of five minutes. The adjuster still makes every substantive decision (coverage, value, whether anything warrants a closer look). The model gets the file in front of them faster and in a consistent shape.

Where this is not a drop-in replacement

An LLM summarizing and triaging claims is not a fraud detection system, and using it as one is a mistake. Insurance fraud detection relies on statistical models trained on historical claim and payout data, known-pattern matching and specialized tooling, often under its own regulatory oversight. An LLM asked "does this claim look suspicious" gives a confident answer from surface features of the text, without the base rates a real fraud model has. Use it to flag a claim for a human to look at more closely, or as one input alongside your actual fraud detection tooling, never as the tool that clears or flags a claim on its own.

One Spark for most claims teams, a linked pair for heavy files

Claim files range from a short auto claim to a long medical claim with imaging reports. A single Spark's 128 GB of unified memory runs a large instruct model with room for a long context on the bigger files, and a $0.79/hour on-demand instance covers most claims teams. A linked two-node spark-2x cluster at $1.79/hour is worth considering when volume and document length both run high, such as a health insurer processing imaging reports with medical claim narratives.

Related pages

First top-up: pay $20, get $30 in credit

Triage claims without a third-party AI API.

A dedicated DGX Spark in EU-Central, $0.79/hour, for claims triage and summarization on a machine assigned only to you.

Your data stays on a machine assigned only to you. GPUwerk processes it under its published DPA.

Create account Private LLM hosting