AI & Data Transparency

Effective Date: January 1, 2026

1. Utah AI Policy Act Disclosure

Bot Disclosure:

When you use ProCode AI, you are interacting with Generative Artificial Intelligence technology, not a human being. The codes and summaries are generated by probabilistic models.

This disclosure is provided in compliance with the Utah AI Policy Act (2024) and applicable state laws regarding automated systems in regulated occupations.

2. How We De-Identify Your Data

For ProCode AI, we process your clinical notes to create a library of anonymous coding examples. We use a combination of methods to ensure privacy:

AI

LLM-Based Redaction (High Recall)

We utilize advanced Large Language Models (LLMs) specifically tuned for "high-recall" detection. This means the model is biased toward over-redacting to ensure no identifiers are missed. We target the 18 identifiers specified under the HIPAA Safe Harbor rule (45 CFR § 164.514(b)(2)).

db

Snippet Retention

Once data is fully de-identified, it is no longer Protected Health Information (PHI). We store these anonymous snippets (e.g., "patient presents with acute cystitis..." mapped to "N30.00") to retrain and improve our models. We do not retain the raw, original note after the session is closed.

3. Third-Party Processing

ProCode utilizes enterprise-grade LLM providers for processing. We have executed "Zero Data Retention" agreements with these providers, ensuring that:

  • They do not use your PHI to train their foundation models.
  • They do not store your request logs after the session is complete.

4. Limitations & Professional Verification

Automated Suggestions vs. Clinical Judgment

ProCode AI utilizes advanced validation layers to ensure that every suggested code exists in the official standards and is grounded in the text of your documents. However, Artificial Intelligence lacks human clinical context.

While we filter out non-existent codes, the AI may still:

  • Misinterpret Context: It may link a valid clinical snippet to a code without recognizing that the condition was "ruled out," "suspected," or part of "family history" rather than a current diagnosis.
  • Generate Flawed Rationales: The logic provided to explain a code selection may sound plausible but be clinically invalid.
  • Omit Information: The AI may fail to capture every relevant diagnosis present in complex notes.

You acknowledge that ProCode is a decision-support tool. You must independently verify every suggestion against the original clinical documentation before billing. You assume full responsibility for the final codes submitted to payers.