Arbital Health

AI agents perform actuarial investigations within a broader financial contract platform. They turn a user task into an analytical plan, select specialized tools, interpret intermediate results, and produce an answer linked to contract economics. Saved Workflows make those analyses repeatable, while Workspaces allow teams to continue examining the findings together. These functions are substantive uses of learned models.

However, the underlying claims ingestion, contract reconciliation, financial records, benchmarks, and human actuarial services also have value independently. The offering therefore combines an important AI analysis layer with established data and financial operations rather than depending on AI for every core function.

A user supplies the task, after which agents can plan and execute multiple analytical steps and adjust their approach as results arrive. Plans, progress, underlying queries, and source data provide review points, and shared Workspaces remain editable. Actuaries and finance professionals retain judgment about whether a variance is meaningful and whether a number is ready for decision-making.

Fully proactive analysis of an entire book of business is described as future work rather than the current default. Numeric escalation thresholds, precise stopping conditions, and a complete override procedure remain unspecified, although the current workflow establishes a substantive human oversight role.

The current agents combine several foundation models with actuarial reasoning and tools designed for tasks such as benchmark comparison and contract-term lookup. They plan analyses, select tools, and use intermediate results to decide subsequent steps. This gives a useful architectural explanation of how the conversational interface reaches financial conclusions. It does not identify the actual model families, suppliers, deployed versions, training sources, or model-change schedule.

The distinction matters because a change in the language model can alter task interpretation even when the underlying financial data stays constant. Actuary-authored tests are described, but their coverage and release thresholds are not fully specified.

The architecture uses several foundation models assigned to different actuarial tasks, together with specialized analytical tools and hosted infrastructure. The privacy notice also describes service providers for hosting and chat communications. Actual model providers, the complete inference route, and provider-specific retention or training terms remain unnamed. Marketing and analytics suppliers listed elsewhere in the policy do not identify who processes a customer's actuarial prompts. This leaves only a partial view of the model supply chain.

Arkos Health and SonarMD describe using Arbital, and the homepage includes a 6:1 return claim. SonarMD specifically connects Flex with analysis of its inflammatory bowel disease data. These are named adoption signals, but a defined evaluation period, comparator, and reproducible calculation for the return are unclear. An older anonymous surplus-payment case combines negotiation, consulting, and reconciliation work, so its result cannot isolate the contribution of current AI agents.

Likewise, more than 1,500 platform contract reconciliations measure operating experience rather than agent accuracy. Current agent outcome evidence remains distinct from testimonials and the financial performance of earlier services.

The privacy notice permits service improvement and creation of deidentified or aggregated information, with a commitment against reidentification except where legally permitted or required. It does not specify a patient-data training prohibition, retention window, or detailed deidentification method. On analytical safety, actuaries write expected answers for test questions, and specialized tools reduce reliance on newly generated database queries.

Results can be traced through their source data and calculations. Those practices help expose errors, but model-specific hallucination acceptance thresholds and a complete response process for harmful analytical mistakes remain unclear. General security incident handling does not establish the same thing.

The AI platform claims HIPAA compliance and operates on the same infrastructure that holds customer contract and population data. A substantive privacy notice covers account information, uploaded information, platform workflows, reporting interactions, communications, and customer access to associated files. It also explains service-provider sharing and individual privacy rights. A readable BAA defining applicable products, service tiers, and subcontractor obligations remains unavailable.

The compliance claim and service privacy notice therefore provide a stated healthcare data posture without establishing the full contractual chain for PHI. Security certifications address a related assurance question but do not replace those business associate terms.

Arbital reports HITRUST i1 certification and a 2026 SOC 2 Type 2 audit for its platform. Its trust center lists supporting assurance documents, but access to the reports requires a request. These are recognized security assurance claims connected to the customer service, rather than merely descriptions of encryption or access control. The full audit boundaries, exceptions, covered infrastructure, and detailed period of examination remain gated. As a result, the visible claims establish a stated certification and audit posture while leaving direct confirmation of the current agent-processing scope dependent on the underlying documents.

The principal functions concern contract economics, claims utilization, population spending, and financial risk. Users can investigate high-cost members or readmission-related patterns, but the intended workflow remains operational and actuarial analysis for payer and provider teams. The platform does not claim authority to diagnose a patient or prescribe a treatment.

References to improving care or directing resources describe business and population-management goals rather than a medical device indication. No medical device clearance is claimed for the current platform. Financial and clinical staff interpret the analyses within their own responsibilities, including decisions about where to focus care-management resources.

Actuaries define analytical methods, select expected answers for test questions, and test reusable Workflows and Workspaces. The current architecture also uses task-specific tools and several foundation models assigned to different jobs. This is a concrete development and evaluation practice. A formal governance framework, customer review artifact, documented release acceptance criteria, and comparative subgroup evaluation remain unspecified.

Illustrative age and gender charts show how populations can be analyzed; they do not measure model fairness. General benchmark figures discussed in the company's commentary describe external systems, not Arbital's own accuracy. The current assurance is therefore narrower than a formal, independently inspectable governance program.

Saved queries, editable Workspaces, and traceable calculations provide practical ways for teams to inspect and challenge a result. Users can examine the members, claims, processing steps, and contract terms behind a number, then revise the analysis or obtain actuarial interpretation. These controls can help catch an incorrect conclusion before a financial decision. They do not establish an output warranty, reimbursement for a contract loss, or a contractual correction obligation.

Privacy access and correction rights concern personal information rather than a dedicated appeal of an actuarial recommendation. The available remedy is therefore principally review and operational correction within the workflow.

Flex accepts medical and pharmacy claims, eligibility, revenue, and custom reference files, including provider rosters, engagement information, and attribution logic. Data quality review precedes analysis, and the main platform supports ingestion and reconciliation across different file types. These are concrete data intake capabilities for actuarial work. A named EHR connector, interface standard, supported clinical-system version, and verified production exchange remain unspecified.

Member-level analysis therefore does not establish that results flow automatically back into a clinical record. The described workflow centers on uploaded and reconciled financial and population data rather than demonstrated bidirectional EHR integration.

The main product operates as a hosted SaaS platform, and Flex lets customers upload files without a full implementation project. The privacy notice allows personal information to be stored and processed in the United States and elsewhere worldwide. A committed customer data region, tenant separation design, backup map, and boundary for international model processing remain unspecified. The cloud delivery model is clear, but it does not establish a residency guarantee for the complete analysis workflow.

The commercial offering includes the main contract management platform, Flex for direct data uploads, and actuarial services. Flex can support population analysis without a formal risk contract, making it a distinct entry route for earlier-stage users. Customers can also combine software with specialist interpretation and project work. Public subscription amounts, charging units, implementation fees, minimum commitments, and the cost of adding AI functions remain unspecified.

Self-service file upload describes how customers use the product, rather than a self-service purchasing or pricing model. A buyer can identify the available service categories but cannot calculate a comparable initial budget before commercial engagement.

The platform serves payer, provider, and care-enabler teams managing financial risk across populations and contracts. Finance, clinical, and actuarial staff can share analyses of cost, utilization, membership, and condition-specific spending. SonarMD provides a named specialty example involving inflammatory bowel disease, while Arkos Health supplies another organizational adoption signal.

Flex broadens the setting to populations without an established risk contract, using uploaded files and national benchmarks. This supports use across more than one customer type and analytical setting. Measured outcomes across the entire advertised range remain limited, particularly for the newer agents rather than the underlying reconciliation and advisory services.