AI Disclosure

CUBE OneCare embeds AI in the following product surfaces:

• Clinical documentation assistance. Generating draft SOAP notes, summarizing encounter transcripts, and suggesting ICD-10/CPT codes that providers review and accept before storing.
• Patient-facing triage and self-service. Symptom-check flows and the "CUBE Assist" chat widget answer general questions, link to appointments, and route to human support for anything that looks clinically urgent.
• Operational AI. Duplicate-patient detection, coding-error triage, schedule-optimization suggestions, and anomaly detection in usage metrics.
• Infrastructure AI. Automated log summarization, incident triage, and on-call escalation routing.

The platform uses CUBE Mind, a proprietary orchestration layer that combines eligible foundation models with internally trained specialist models. Vendor allocation is not exposed in public product surfaces; authorized enterprise and regulatory reviewers can request provider-level detail under NDA. All AI inference calls route through our inference gateway so we can enforce:

• Strict prompt-construction and safety-filter policies.
• Row-level audit logging of every inference, tied to the requesting user and purpose.
• Contractual opt-out from provider training on submitted content — no customer PHI or PII leaves CUBE OneCare's control for the purpose of model training.

CUBE Collective LLC does not train general-purpose machine-learning models on PHI or on customer content. Specifically:

• Customer content is not used to train any third-party foundation model we integrate with. This is enforced contractually with each foundation-model provider and technically via the training-opt-out flags of each API.
• Where we build internal specialist models (for example, a coding-error classifier), we train on either de-identified data sets under HIPAA § 164.514(b) Safe Harbor, fully synthetic data, or data for which explicit written consent has been obtained for the training use case.
• We do not sell training data to third parties.

AI outputs that touch clinical decisions are surfaced as suggestions for licensed clinicians to accept, edit, or reject. The platform enforces this oversight in several ways:

• No AI-authored clinical note is stored in the patient record until a provider has signed it.
• Triage flows that detect red-flag symptoms break out to human staff rather than continuing with an automated answer.
• We run model-quality evaluations against curated benchmark sets (including a hallucination suite focused on medication dosing and adverse-event recognition) before each release of a model update. Results are documented for the customer Trust Center.

AI outputs can be wrong or incomplete. We ask users to:

• Treat every AI-produced recommendation as advisory — always verify against primary sources and clinical judgement.
• Never rely on the CUBE Assist for medical emergencies.
• Report hallucinations or unsafe outputs so we can update evaluation sets.

Reports should go to [email protected].

Every AI-assisted feature is labeled inside the product; users can always see when they are reading AI-generated content. Administrators can disable specific AI features at the tenant level for compliance reasons. Individual users can opt out of AI-assisted documentation for their own encounters at any time from the account settings.

We monitor regulatory developments including the FDA's AI/ML Software-as-a-Medical-Device guidance, the EU AI Act, and state-level acts on automated decision-making. The AI features in CUBE OneCare are currently positioned as decision-support tools, not as regulated medical devices, and are documented accordingly. If a feature is reclassified by regulators, we will update this disclosure and customers prior to enforcement dates.