HIPAA's Security Rule and Privacy Rule together establish the minimum technical and administrative safeguards required to protect electronic protected health information. The introduction of AI agents into clinical workflows — agents that access EHR systems, scheduling platforms, and patient records — creates new risk categories that the existing HIPAA risk analysis obligation requires covered entities to assess and address.
Section 164.308(a)(1) requires covered entities and business associates to conduct an accurate and thorough assessment of potential risks and vulnerabilities to the confidentiality, integrity, and availability of ePHI. This is not a one-time exercise — it is an ongoing obligation that must be updated when new technology, including AI agents, is introduced into the environment. The HHS Office for Civil Rights has consistently cited inadequate risk analysis as the root cause in enforcement actions.
Section 164.312 requires technical safeguards including access controls limiting ePHI access to authorized persons and software, audit controls that record and examine access activity, and transmission security ensuring ePHI is not intercepted in transit. AI agents that query EHR systems, retrieve patient records, or initiate scheduling actions are subject to these technical safeguard requirements — and most organizations have not extended their audit and access-control infrastructure to govern agent behavior.
Any vendor that creates, receives, maintains, or transmits ePHI on behalf of a covered entity is a business associate and must execute a BAA. A surveillance vendor that processes communications containing PHI — or an AI governance tool that intercepts agent actions involving patient data — is a business associate. The architecture of the tool determines whether PHI is actually transmitted: if inspection happens inside the covered entity's own environment, the ePHI is never transmitted to the vendor.
Healthcare organizations are deploying AI agents at speed: patient intake bots, scheduling agents, clinical decision-support tools, and care-coordination assistants that reach directly into EHR systems and patient records. These agents operate faster than any human review and against no real-time control. When an agent retrieves or outputs PHI improperly — sending it to an unauthorized recipient, including it in a response that is logged outside a secure environment, or sharing it across a boundary that violates the minimum-necessary standard — the violation has already occurred by the time it appears in an audit log.
HIPAA's risk analysis obligation requires organizations to assess these risks before deploying AI systems that touch ePHI. A risk analysis that identifies the risk without implementing a technical control to address it is an incomplete risk management program. The OCR's enforcement history shows that inadequate risk analysis, not merely breach of PHI, is independently actionable.
The governance challenge is compounded by the architecture of most AI surveillance tools: they process communications in a vendor cloud, which means PHI in those communications is being transmitted to a business associate. In-tenant inspection removes this exposure structurally rather than contractually.
Compiled deploys inside your Azure tenant or on-premises environment. Every communication and every AI-agent action involving patient data is inspected inside your boundary. PHI is never transmitted to Compiled or to any vendor-operated infrastructure.
OCR investigations and HIPAA audits typically focus on whether the covered entity had appropriate safeguards in place and whether those safeguards were operating. Compiled produces a structured record from inside your own environment that documents both.