HIPAA applies to an AI agent the moment it reads or writes protected health information for a covered entity or business associate. Its uses must fit treatment, payment or health care operations, or have patient authorisation. Its access needs security rule safeguards. Anything it writes into a record must be accurate and traceable, and its vendors need business associate agreements.
What does HIPAA require?
HIPAA protects individually identifiable health information held or transmitted by covered entities and their business associates. In substance it works through three sets of rules.
Privacy. Protected health information may be used and disclosed without the patient's authorisation for treatment, payment and health care operations, and for certain other permitted purposes. Other uses generally need written authorisation. Patients have rights to access their records, request amendments, receive an accounting of certain disclosures and complain about privacy practices.
Security. Electronic protected health information must be protected through administrative, physical and technical safeguards. That starts with a risk analysis and includes access control, audit controls, integrity controls, authentication of people and systems, and transmission security.
Breach notification. Breaches of unsecured protected health information must be notified to affected individuals and the federal regulator, and in larger cases to the media, within set deadlines.
Who does it apply to?
Covered entities, which are health plans, health care clearinghouses and health care providers that conduct standard electronic transactions, and their business associates. Whether a particular agent deployment brings you in scope is covered in do AI agents need to comply with HIPAA. A wellness app that is not working for a covered entity may fall outside HIPAA while still facing state law and consumer protection rules.
What changes when the system is an agent that reads and writes?
Agents in health settings tend to do four jobs, and each hits HIPAA differently.
| Agent job | HIPAA point it touches | Main control |
|---|---|---|
| Answering patient questions about appointments, bills or results | Disclosure to the right person; identity and personal representatives | Verified identity before any PHI is read |
| Summarising records for clinicians or claims staff | Minimum necessary; accuracy | Access scoped to the task; summary marked as derived |
| Writing to the record: notes, codes, prior authorisation forms | Integrity of PHI; amendment rights | Drafts held for a person; every write attributed |
| Sending PHI to outside parties: payers, labs, patients | Permitted disclosure; transmission security | Destinations limited to approved recipients |
Writing is the job that most needs care. An agent that drafts a clinical note or assigns a billing code is creating PHI that others will rely on. The security rule's integrity standard expects you to protect PHI from improper alteration, and the privacy rule gives patients a right to request amendment. Both need a record of what the agent wrote, from which sources, and who approved it.
Patient-facing agents add a second risk. A parent, carer or relative may contact the agent on a patient's behalf. HIPAA's rules on personal representatives and disclosures to family members require judgement. An agent should not make those calls; it should route them to a person.
How do you meet HIPAA for agents?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Permitted use or disclosure | Each agent capability mapped to treatment, payment, operations or an authorisation | Capability register with HIPAA basis |
| Minimum necessary | Access limited to the fields the task needs | Permission map per capability |
| Risk analysis | Agents included in the security risk analysis | Risk analysis entries |
| Access control and authentication | Unique agent identity; patient identity verified before access | Identity register; verification logs |
| Audit controls | Every read and write by the agent logged | Audit logs retained per policy |
| Integrity | Agent writes drafted for approval, or attributed and reversible | Approval records; change history |
| Business associate agreements | Agreements with the agent platform and its sub-contractors handling PHI | Signed agreements |
| Patient rights | Agent records searchable for access and accounting requests | Completed request files |
| Complaints | Privacy complaints raised to an agent reach the privacy officer | Complaint routing records |
| Breach readiness | Agent incidents assessed under breach rules | Incident and assessment records |
Complaints deserve a note. Patients increasingly raise privacy concerns in the same channels the agent covers. The agent must recognise a complaint and pass it on, not resolve it; our explainer on what complaint handling is sets out the basics. Access scoping is covered in detail in HIPAA minimum necessary and AI agent access.
What will an auditor or investigator ask?
- Which agents access PHI, and under which permitted purpose?
- Is the agent in your risk analysis, and what risks did you find?
- Show the audit trail for one patient: what did the agent read and write?
- Who approves what the agent writes into the record?
- How does the agent confirm identity, and what does it do with a caller claiming to be a relative?
- Where are the business associate agreements for the platform and model provider?
What does HIPAA not cover?
It does not cover health data held by organisations outside its scope, such as many consumer apps. It does not regulate the clinical quality of an agent's output; that is a matter for clinical governance and, for some tools, medical device rules. It does not cover financial data, which falls under GLBA; see does GLBA apply to AI agents handling customer data. Outside the US, other regimes apply, such as Canada's federal privacy law in PIPEDA and AI agents handling customer data. For the broader map, see AI agent regulation, mapped to controls.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Questions people ask
Can an agent share test results directly with a patient?
HIPAA allows disclosure to the patient, and patients have a right of access. The question is clinical: some results should come with a conversation. Decide which results an agent may release, and hold the rest for a clinician.
Does de-identified data fall outside HIPAA?
Yes, if it is properly de-identified under HIPAA's methods. Free text from agent conversations is hard to de-identify fully, because it carries names, dates and details. Do not assume a redacted transcript qualifies.
Is an agent's summary of a record itself PHI?
Yes. A summary that identifies a patient and describes their health or care is PHI, and it is subject to the same protections as the source record.