Compliance Guide

    Texas SB 1188 AI Requirements for Diagnosis and Treatment Recommendations

    As of November 2026, the SB 1188 duties that most directly reach AI used to support diagnosis and treatment recommendations are its electronic health record requirements. A covered entity must ensure that EHRs under its control containing patient information are physically maintained in the United States or a U.S. territory, including when held by third-party or cloud providers. It must also ensure that Texas residents' EHR information is accessible only to people who need it for treatment, payment or health care operations. Applying these duties to AI workflows is interpretation, and this guide labels it that way.

    Gap assessment for current AI deployments

    • List every store of EHR-derived patient information in each AI workflow, including logs, vector stores and vendor subprocessors, and record its physical location.
    • Review cloud, model-hosting and subcontractor agreements for data-location commitments covering EHR information.
    • Map AI service identities and users to treatment, payment and health care operations purposes, and test whether any can retrieve more than the human user could.
    • Check whether voting-related or voter-registration data can enter AI inputs, context stores or logs.
    • Confirm you can produce, for a named encounter, the policy that applied and what the AI system did with patient data.
    • Assign an owner for the final text, operative dates, penalty provisions and any rules from the Texas boards or the Texas Department of Insurance.

    What SB 1188 requires

    SB 1188 of the 89th Legislature, Regular Session, is titled as relating to electronic health record requirements. It amends the Health and Safety Code to place duties on a covered entity with respect to EHRs under its control. Three duties matter for clinical AI.

    First, EHRs that contain patient information must be physically maintained in the United States or a U.S. territory. The location duty extends to EHRs stored by a third-party or subcontracted computing facility or a cloud computing provider, and to EHRs stored using technology through which patient information may be electronically retrieved, accessed or transmitted.

    Second, EHR information of Texas residents, other than open data, must be accessible only to individuals who need it to perform duties within the scope of their employment related to treatment, payment or health care operations.

    Third, the bill as analyzed by the Senate Research Center prohibits EHR platforms from collecting and retaining information about a person's voting history or voter registration status, and from collecting personal information used to help a person register to vote or execute a mail-in ballot.

    The bill defines a health care practitioner as an individual licensed, certified or otherwise authorized to provide health care services in Texas. It also includes a provision for a Texas board and the Texas Department of Insurance to adopt rules as necessary to implement the chapter. A later version of the bill is titled as authorizing a civil penalty. Counsel should confirm the penalty, who enforces it and what conduct it covers before any risk rating relies on it.

    Where AI enters the picture

    An AI tool that supports diagnosis or treatment recommendations typically receives patient information in a prompt, retrieves it as context, stores it in a vector index, sends it to a hosted model, and writes it to logs. Each of those is a place where EHR-derived information may be held or transmitted.

    The location duty is written around EHRs under a covered entity's control, including those held by third-party and cloud providers. It is a reasonable reading that AI vendors, model hosts and logging stores holding patient information fall within the duty. That is interpretation, and the answer under the enacted text is a question for counsel.

    The access duty is role-based. It turns on whether a person needs the information for treatment, payment or health care operations. Interpretation: an AI service should not widen that access. If a model or agent can retrieve records that the clinician using it could not, the AI path becomes a way around the access rule, regardless of how well the EHR itself is permissioned.

    The voting-information provision is addressed to EHR platforms. Interpretation: if the final text keeps it, inputs to AI workflows are one place to show that such data is not being collected or retained.

    Who owns each obligation

    The duties described above attach to the covered entity, not to a specific tool. A practitioner using an AI system does not discharge them by clicking through a screen, and a vendor does not remove them from the covered entity by hosting the data. Separate the work along these lines.

    The organization owns the inventory of systems holding EHR-derived information, the contractual commitments from cloud, model-hosting and subcontractor providers about data location, and the definition of roles and purposes for access. Practitioners and clinical leaders own how AI output is used in care, and the organization's own policies on that sit outside what runtime controls can decide.

    Supporting systems can enforce and evidence what the organization has decided. They can apply an access rule to each request, keep patient data from reaching a model or tool that should not receive it, and retain a record of the decision. They cannot establish that a contract term is accurate or that a purpose was legitimate.

    Before building a control matrix, confirm how the final text divides duties between a covered entity and a health care practitioner, since the two terms come from different parts of the bill's history.

    Requirement-to-control mapping

    Each row states the obligation, the operational question it raises, and the evidence that answers it. Rows marked as interpretation are mappings to AI workflows, not statutory text.

    • EHRs kept in the U.S. or a U.S. territory: Question: where is each AI-related store of patient information physically maintained? Control: inventory of prompts, context, model hosts, logs, vector stores and subprocessors, plus contract review. Evidence: location record per store.
    • Third-party and cloud storage covered: Question: do provider agreements keep EHR information in scope of the location duty? Control: contractual data-location commitments. Evidence: executed terms and configuration records.
    • Access limited to treatment, payment, operations: Question: can an AI service or user reach records outside that purpose? Control: least-privilege scoping of AI identities, enforced per request (interpretation). Evidence: access decision records.
    • No voting-related data collected or retained: Question: can such data enter AI inputs or logs? Control: input filtering by redaction or blocking (interpretation). Evidence: enforcement records showing the policy and its outcome.
    • Rules and penalty exposure: Question: what rules apply and who enforces them? Control: a named owner tracking final text and board or Texas Department of Insurance rules. Evidence: dated internal determinations.

    Runtime enforcement and evidence

    Written policy and contract terms do not show what an AI system did with patient data in a given encounter. A regulator or auditor asking that question needs a record of what the system was permitted to do and what it did.

    Policy enforcement in the execution path is one way to produce that record. Trussed enforces policy on individual interactions and tool calls across models, agents, tools and MCP servers: allow, redact, block, rate limit, or escalate for human approval. Applied to this statute, a request that would send patient information to a destination outside approved scope can be blocked, identifiers or voting-related content can be redacted before a model sees them, and an unusual access can be escalated to a person for approval. These are design options. The bill does not itself require these controls.

    The evidence comes from the same enforcement point. Trussed generates audit-ready evidence from enforcement, including what policy applied and what the system did, mapped to frameworks such as HIPAA. It can be self-managed in the customer's cloud with configurable retention and data residency, which can help an organization keep the evidence itself in the location it chooses. Enforcement does not replace contract review or a confirmed location inventory. It shows that the access and handling rules the organization set were applied at runtime, not reviewed after the fact.

    From statutory duty to operational evidence

    Location

    Where EHR-derived patient information is physically maintained, including cloud and subcontractor stores.

    Access

    Who and what can reach Texas residents' EHR information, and for which purpose.

    Enforcement

    Policy applied to patient data in each AI request before it reaches a model or tool.

    Evidence

    A retained record of what policy applied and what the system actually did.

    Show what your AI was allowed to do and what it did

    Talk through how enforcement in the execution path and audit-ready evidence support the access and handling duties in your clinical AI workflows.

    Talk to an Expert