AI security and governance

Last updated: Jun 2, 2026
HEALTH TECH VENDOR
HCO

This article details how we secure and govern the Redox AI Assistant Suite.

The AI Assistant Suite includes three AI-powered assistants (the “AI assistants”) embedded in the Redox dashboard: Config Modifier Assistant, Log Summary Assistant, and Payload Summary Assistant. The features are designed to enhance integration management workflows, including self-service capabilities, while maintaining the rigorous standards required for Protected Health Information (PHI). 

The AI assistants inherit Redox platform security, compliance, and privacy protections, and we ensure sensitive data, such as PHI, is never used to train underlying AI models.

Data classification & handling

We distinguish between two types of data powering the AI assistants:

1. Usage data

  • What it covers: Metadata from AI invocations, such as timestamps, (when it was used), user IDs (who used it), user feedback, and operational details like response times, success rates, or errors.
  • Purpose: To monitor usage trends, enforce rate limiting, and optimize the user experience through built-in feedback mechanisms (e.g., thumbs up/down).
  • Storage: Usage data is stored indefinitely.
  • Training: AI assistants enabled in the Redox dashboard, may use your usage data for training purposes. This data may be anonymized and aggregated with data from other Redox customers to train and improve the AI assistants.

2. Prompt data

  • What it covers: Prompts, prompt context (e.g., log data), and AI outputs. Prompt data may contain sensitive information, including PHI.
  • Purpose: Troubleshooting technical issues and providing support for AI failures or problems that users may encounter while using the AI assistants.
  • Storage: We maintain a temporary log of AI interactions (e.g., question asked and the answer given). The data is stored in a PHI-safe datastore and permanently deleted after 60 days in accordance with our standard 60-day data retention policy. Stored data is strictly restricted to Redox team members authorized to view PHI, and all access to prompt data is actively audited. After PHI is removed, prompt data is stored indefinitely.
  • Training: Once all PHI is removed, prompt data may be used for training purposes. To be clear, PHI will NOT be used to train any features within the AI Assistant Suite.

No clinical data use, just contextualized prompts: The AI assistants are designed solely to support technical configuration tasks in the dashboard. The AI assistants do not learn from, generate, modify , or create PHI. They should also not be used as clinical diagnostic tools. All PHI processing remains governed by standard Redox integration protocols.

Secure, private AI inference: Our AI assistants leverage a fully HIPAA-compliant inference engine hosted on AWS Bedrock. This environment is purpose-built to securely handle PHI and strict data boundaries are enforced, meaning sensitive data, including PHI, is never used to train or improve the underlying AI models.

Technical safeguards & encryption 

Data used to power our AI assistants is fully protected by Redox’s enterprise-grade platform security and privacy safeguards to prevent data misuse and exposure. Governed by HITRUST controls and a SOC 2 Type 2 framework, these features leverage built-in security controls, including end-to-end HTTPS encryption, private network hosting, robust access management, and proactive 24/7 account monitoring.

  • Role-Based Access Control (RBAC): The AI assistants inherit the existing permissions of the logged-in user, and every action is scoped to a user’s existing Redox role. Our built-in safeguards ensure security and governance is enforced by the platform, not the user.
  • Data security: All data, including every prompt and response, is fully encrypted both in transit and at rest to prevent unauthorized access at every stage of the workflow.
  • No passive AI processing: Our AI assistants are invocation-based and only accesses the specific data points required to answer a user’s direct request. AI assistants in data processing only run when a user explicitly needs it and agrees to it. AI assistants are only active when a user invokes them.

Is my data used to train the AI assistants?

As detailed above, your interactions themselves (usage data and prompt data) may be used to train the AI assistants, but PHI will NEVER be used to train the AI assistants. Any data that could possibly contain PHI is stored in a separate, restricted, and auditable environment, which is subject to our standard data retention policy. Any PHI that may be contained within a prompt is manually replaced with synthetic data prior to being exposed to an environment used for AI training.

Are the AI assistants compliant with HITRUST and other regulations?

Yes. Our entire AI Assistant Suite was designed with all applicable regulatory requirements and with the HITRUST AI security controls in mind.

What underlying AI models do the AI assistants use?

Our AI assistants use the latest available model that is compliant with all applicable regulatory frameworks and is governed by Redoxs strict BAA requirements.

Can I disable the AI features if I don’t want to use them anymore?

Yes. The AI assistants can be turned on or off by an organization owner at any time in the Redox dashboard under Organization Profile > Settings.

Reliability & accuracy

AI accuracy: When using the AI assistants, the user who invokes the AI remains the final authority. AI can occasionally produce“hallucinations” or inaccuracies, and it is the user’s sole responsibility to review and approve the content before it’s applied to a configuration or used in a decision-making process. After a response is generated, a disclaimer indicates that the output was generated using AI.

FHIR® is a registered trademark of Health Level Seven International (HL7) and is used with the permission of HL7. Use of this trademark does not constitute an endorsement of products/services by HL7®.