Hoag Health - AI Transparency

Hoag Health AI Transparency

At Hoag, we are committed to using technology thoughtfully and responsibly to support safe, high‑quality, and compassionate care. This page explains how Hoag uses artificial intelligence (AI), including generative AI, in certain digital health services and how we protect patient trust, privacy, and clinical judgment.

How Hoag Uses Artificial Intelligence

Hoag uses AI‑enabled tools to support—not replace—clinical care. These tools are designed to help organize information, improve efficiency, and enhance patient experiences.

AI may be used to assist with:

• Patient intake and symptom collection

• Summarizing patient‑provided information for clinical review

• Care coordination and visit routing

• Drafting documentation, summaries, and after‑visit materials

• Transcription and summarization of telehealth visits

AI systems do not independently diagnose medical conditions or determine treatment. All care decisions are made by licensed health care professionals.

Human Oversight and Clinical Responsibility

All clinical decisions at Hoag are made by licensed providers.

AI‑generated outputs are used only as clinical support tools. Clinicians review relevant information, apply professional judgment, and determine the appropriate diagnosis, treatment plan, and level of care.

Use of AI Systems

Hoag designs, codes, and operates AI‑enabled services for use by patients and the public, including Ask Hoag, using third‑party AI technologies within secure, HIPAA‑compliant environments.

Hoag’s activities include:

• Defining the role, scope, and purpose of AI within specific clinical and administrative workflows

• Writing instructions that guide how AI responds within those workflows

• Implementing safety controls and response guardrails

• Integrating approved reference and contextual data to support real‑time AI interactions

• Monitoring performance to ensure safety, accuracy, and appropriateness

Model‑Level Training and System Configuration

Hoag does not perform model‑level training, retraining, or parameter fine‑tuning of generative AI models, meaning Hoag does not alter the underlying model architecture, weights, or learning algorithms.

Hoag does, however, design and operate AI‑enabled services by configuring how generative AI capabilities are used within specific clinical and administrative workflows. This includes writing system instructions and prompts, implementing response guardrails, establishing workflow logic, and incorporating approved reference and contextual data to guide AI outputs for defined use cases. These activities shape how AI functions within Hoag‑operated services and constitute part of Hoag’s development of those services, but they do not involve retraining or modifying the underlying AI models themselves.

Patient information is not used to retrain, update, or fine‑tune the underlying AI models.

Safety Controls and Guardrails

Hoag implements multiple layers of safeguards to ensure AI is used safely and appropriately, including:

• Purpose‑limited instructions designed specifically for health care use cases

• Guardrails that restrict AI outputs to supported administrative and clinical support functions

• Controls that escalate interactions to licensed clinicians when clinical judgment is required

• Ongoing monitoring and review to identify and address potential risks

Training Data and Datasets

(California AB 2013 Disclosure)

The AI systems used by Hoag rely on models and components developed using large datasets. The following disclosures are provided in accordance with California Civil Code § 3111.

Whether Datasets Were Purchased or Licensed

The datasets used to train the generative AI systems underlying Hoag’s AI‑enabled services include datasets that were purchased and/or licensed by the AI system providers, as well as publicly available data and data created by human trainers. Hoag itself does not purchase or license datasets for the purpose of training generative AI models.

Data Sources and Origins

Training datasets used by AI system developers may include:

• Publicly available data

• Licensed data sources

• Data created by human reviewers or trainers

Hoag does not provide patient medical records or identifiable patient data for the purpose of training AI models.

Purpose of the Datasets

Training datasets are used to:

• Enable general language understanding

• Support summarization and contextual responses

• Improve system safety and reliability

Dataset Size

Training datasets typically consist of large volumes of data, often including millions to billions of data points.

Types of Data

Depending on the dataset, data may include:

• Textual information

• Structured and unstructured language data

• Artificially generated (synthetic) data

Copyrighted and Licensed Content

Datasets may include licensed, copyrighted, or public‑domain content. Use of such content is governed by applicable legal and licensing requirements.

Personal and Sensitive Information

• AI systems used by Hoag are not trained using identifiable patient medical information

• Training datasets used by AI system developers may include

  aggregate consumer information

  , meaning information that relates to groups or categories of consumers and is not reasonably linkable to any individual or household

• Aggregate consumer information is distinct from identifiable or de‑identified individual records

• During care delivery, any patient data processed by AI is handled in accordance with

  HIPAA and applicable privacy laws

• Patient data is used only for treatment, health care operations, and quality improvement as permitted by law

Aggregate Consumer Information

Training datasets used by AI system developers may include aggregate consumer information, as defined under California law, which relates to groups or categories of consumers and is not reasonably linkable to any individual or household.

Data Processing and Modification

Before being used for training, datasets may undergo processing steps such as cleaning, filtering, de‑identification where appropriate, and formatting to improve accuracy, safety, and reliability.

Time Period of Data Collection

(§ 3111(a)(10))

Training datasets used to develop the generative AI models underlying Hoag’s AI‑enabled services were collected over multiple years prior to Hoag’s deployment of those services. Based on publicly available information and information provided by AI system developers, dataset collection spans several years and may be ongoing as developers continue to update and refine their systems. Hoag does not control or independently verify the specific dates of data collection.

Dates Datasets Were First Used

(§ 3111(a)(11))

Training datasets were first used during the initial development of the underlying generative AI models by their developers, prior to Hoag’s use of those models within its services. Based on publicly available information, such use began several years before Hoag’s deployment of AI‑enabled services. Hoag does not determine or control the timing of dataset use during model development.

Synthetic Data

Some AI systems use synthetic data to test system behavior, improve performance, and reduce reliance on real‑world sensitive information. Synthetic data does not represent real patients.

Transparency, Choice, and Patient Support

Hoag provides notice when generative AI is used in patient‑facing digital experiences. Patients may request human support or in‑person care when appropriate. For emergencies, call 911 or go to the nearest emergency department.

Governance and Ongoing Review

Hoag regularly reviews its AI‑enabled services to ensure alignment with patient safety, clinical quality standards, privacy requirements, and evolving legal and regulatory guidance.

Learn More

If you have questions about Hoag’s use of artificial intelligence, please speak with your care team or contact Hoag Health.