Key facts
| Access control | Scoped API keys with rotation; enterprise SSO and RBAC options |
| Deployment | Cloud, VPC, on-prem or air-gapped to match policyholder rules |
| Auditability | Request, model and response logging for claims and decision review |
| Data grounding | Embeddings and RAG are live for policy wording and claims guidance |
| Structured output | JSON mode returns claim and intake records in fixed schemas |
| Models | 30+ models behind one OpenAI-compatible API |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Audio, images, files, batch and fine-tuning are coming soon |
TL;DR
- Classify claims, underwriting and servicing data separately before pilots.
- Scope keys by function and channel, not one key per insurer.
- Keep a documented human reviewer on any customer-affecting output.
- Log model, sources and reviewer so decisions can be reconstructed.
- Decide residency and retention explicitly for policyholder records.
How it works, step by step
- Inventory AI use cases across claims, underwriting, servicing and broker channels.
- Classify the data each use case touches, including identifiers and financial details.
- Choose deployment per class: region-selected cloud, VPC, on-prem or air-gapped.
- Issue per-application and per-channel keys with rotation and central inventory.
- Define log fields and retention: request ID, model, sources, output, reviewer decision.
- Approve a model allow-list with evaluation evidence and re-test on version changes.
- Document the human review step for any output that affects a policyholder.
Try it yourself
Open the AI API key security checklist →
Policyholder data and decision scope
Insurance data splits into public product information, internal underwriting guidance, policyholder records, claims evidence, and financial or health details. Each class needs its own rule for whether it may leave your boundary, and decision-facing workflows need a higher bar than drafting and summarisation.
Most teams start with policy wording and internal guidance, then extend to claims documents only on private deployments where prompts, attachments and embeddings stay inside the environment.
Keys, roles and least privilege
Issue a distinct API key per application, function and environment. Claims, underwriting and servicing should not share credentials, and broker or partner integrations should have their own keys with tight scopes and clear revocation paths. Store keys in a secrets manager, rotate on a schedule, and connect enterprise SSO and RBAC options so console permissions follow HR records.
Never place policyholder identifiers in prompts where retrieval can supply only the fields a task needs.
Residency, retention and claims audit trails
Decide where processing happens and how long records live. Region selection covers many residency needs; VPC, on-prem and air-gapped deployment supports carriers with stricter mandates. Retention applies to prompts, outputs, logs and retrieval indexes, and claims records usually follow their own schedule that outlasts the AI system.
Log enough to reconstruct a decision: request ID, model and version, retrieved source identifiers, output and the reviewer. See AI audit logs for a schema.
Model governance and reviewer sign-off
Keep an approved model list with evaluation evidence, and re-run evaluations when versions or prompts change. Ground answers in current policy wording with citations so adjusters and underwriters can verify. Any output that affects coverage, pricing or a claim outcome should carry a named reviewer, and the checklist should state which decisions may never be automated.
Honest comparison
| Control area | Plugsky capability | Common gap | Owner |
|---|---|---|---|
| Identity | Scoped keys per function, rotation, SSO and RBAC options | Shared adjuster credentials | Security engineering |
| Data boundary | Cloud, VPC, on-prem or air-gapped deployment | Claim documents leave the carrier | Risk and compliance |
| Retention | Configurable logging of requests and responses | No defined retention | Records management |
| Audit trail | Request, model, source and reviewer logging | Decisions not traceable | Internal audit |
| Model governance | Model choice behind one API with evaluation | Silent model changes | Model risk committee |
| Human sign-off | Citations and structured output for reviewers | Automated adverse decisions | Claims leadership |
Frequently asked questions
Does using Plugsky make us compliant?
No. Compliance is your program. Plugsky provides deployable controls - scoped keys, private deployment, logging - that you document and audit against your own regulatory and policyholder obligations.
Can policyholder data stay inside our environment?
Yes. VPC, on-prem and air-gapped deployments keep prompts, documents and embeddings inside your estate, and the API stays OpenAI-compatible.
What should we log for claims decisions?
Request IDs, model names and versions, retrieved source identifiers, outputs and the reviewer, retained under your policy so a decision can be reconstructed end to end.
Can we automate adverse decisions?
No. Keep a named human reviewer for any output that affects coverage, pricing or a claim outcome, and document where automation is not permitted.
Is fine-tuning available for our wording?
Fine-tuning and batch endpoints are coming soon. Today, use retrieval over approved policy wording with JSON mode for consistent output structure.
How do we handle broker and partner access?
Issue separate scoped keys per partner integration, keep them in a secrets manager, and revoke them contractually and technically when a relationship ends.
Where should a pilot start?
Pilot on public policy wording and internal guidance with the free plan, prove citations and logging, then extend to claims workflows on private deployment.