HIPAA and AI
HIPAA governs the use of PHI in AI models. AI models trained on PHI must comply with HIPAA's minimum necessary standard — only the PHI necessary for the specific purpose should be used. De-identified data (data that cannot be linked to an individual) is not subject to HIPAA and can be used more freely for AI model training. Require AI vendors to document their HIPAA compliance and data de-identification practices.
AI Bias in Health Benefits
AI models trained on historical health data can perpetuate or amplify health disparities. If historical data reflects disparities in care access or quality for certain populations, AI models trained on that data will perpetuate those disparities. Require AI vendors to conduct bias testing — evaluating model performance across demographic groups — and to disclose the results.
AI Transparency
Compliance Note
Transparency in AI decision-making is essential for member trust and regulatory compliance. Members who are denied coverage or directed to specific providers based on AI recommendations have a right to understand the basis for those decisions. Require AI vendors to provide explainable AI — the ability to explain why a specific recommendation was made in plain language. Opaque "black box" AI models are increasingly scrutinized by regulators.