Comparison for healthcare compliance leaders

    TRAIGA vs Texas SB 1188: Which Texas AI rules apply to health systems and health plans

    TRAIGA (HB 149) is a general Texas AI statute with an effective date of January 1, 2026. SB 1188 is a health-record bill that includes provisions on practitioner use of AI in diagnosis and treatment. Neither answers the question on its own. Map each AI use case to the law or laws that reach it, then sort the resulting obligations into those that need runtime enforcement and audit-ready evidence and those that need policy, disclosure or contract terms.

    Evidence an auditor or regulator could ask to see

    • A use-case inventory showing which law or laws you concluded apply, with the reasoning and approver
    • Disclosure language and the points in the patient or member journey where it appears
    • Records showing a practitioner or reviewer saw and acted on AI output where oversight is required
    • Policy definitions for PHI handling, with evidence that redaction or blocking occurred at runtime
    • Retention and data residency settings for AI interaction records
    • Vendor contracts that allocate responsibilities for AI systems you deploy but did not build

    Two laws with different starting points

    TRAIGA is the Texas Responsible Artificial Intelligence Governance Act, enacted as HB 149 in the 89th Legislature. Its bill analysis describes the purpose as protecting public safety, individual rights and privacy while encouraging the safe advancement of AI in Texas. The analysis lists an effective date of January 1, 2026, and the enrolled text includes a monetary penalty provision capped at $100,000. As a general-purpose statute, TRAIGA is framed around the people who develop and deploy AI, not around a single industry.

    SB 1188 starts from a different place. It concerns electronic health records and includes provisions on how practitioners use AI in diagnosis and treatment. Its natural reference point is the clinical record and the clinician, where TRAIGA's is the AI system and the party responsible for it.

    This page is a practitioner's orientation, not legal advice. The enrolled statutory text is the authority, and the details here reflect the position as of November 2026. Because the two laws begin from different places, organizing the comparison by statute tends to hide the overlap. Organizing it by use case exposes it.

    Mapping common healthcare AI use cases to the two laws

    The table below lists the questions that decide which law is likely to be relevant to each use case. The answers depend on your facts and on the statutory text, so counsel should confirm each mapping.

    • Clinical decision support: Is a practitioner using the output in diagnosis or treatment? That is the setting SB 1188 addresses. Separately, who developed and who deploys the system? That is the framing TRAIGA uses. Both laws may be relevant to the same tool.
    • Diagnostic assistance: Same two questions, with more weight on practitioner review. Document who reviews the AI output, when, and what happens if the practitioner disagrees.
    • Patient-facing communications: Does a practitioner use the tool in care, or does the organization deploy it directly to patients? Direct deployment points toward TRAIGA. Messages that enter or draw from the health record raise the SB 1188 question as well.
    • Payer utilization management: Health plans should ask whether automated review involves a practitioner's use of AI in diagnosis or treatment, or is an organizational deployment. The answer determines whether SB 1188, TRAIGA or both are relevant.
    • Record-keeping and summarization tools: Any AI that reads from or writes to the electronic health record puts the record-focused provisions in view. Treat these tools as in scope for assessment until the text says otherwise.

    Where scope is likely to be contested

    The hardest cases sit between the laws. A tool that drafts a patient message from chart data, a payer model that flags claims for clinician review, and a vendor-supplied assistant embedded in an EHR can each be described as a practitioner use, an organizational deployment, or both. Definitions of the covered parties and covered AI systems decide these cases, and they should be read in the enrolled text for each entity type you operate.

    A practical stance is to record the reasoning behind every mapping. For each use case, note which law you concluded applies, which provision you relied on, and who approved the conclusion. Where you could not resolve scope, say so in the record and apply the stricter set of controls. A documented, conservative position is easier to defend than an undocumented assumption.

    Sorting obligations by how they can be met

    Whatever the text requires in disclosure, human oversight, data handling and recordkeeping, each obligation falls into one of two groups. This sorting is a general design principle, not a reading of either statute.

    Some duties can only be met outside the execution path. Disclosure to patients or members is a matter of notice language, interface design and workflow. Vendor responsibilities belong in contracts. Role definitions and training belong in policy. A runtime control cannot satisfy these, though it can help show they were followed.

    Other duties depend on what actually happens during an AI interaction. Human oversight is only demonstrable if the system pauses and routes the decision to a person before the action completes. Data handling is only reliable if PHI is redacted or blocked before it reaches a model or tool, and if data stays where policy says it may reside. A durable record of AI behavior has to be produced at the moment of the interaction, not reconstructed afterward from memory or scattered application logs.

    A review after the fact can show that a problem occurred. It cannot show that a control was in force when the interaction happened. For duties in the second group, that difference matters to an auditor.

    Connecting the Texas duties to HIPAA evidence

    Health systems and plans already operate under HIPAA, including PHI safeguards and audit controls at 164.312(b). Evidence built for those controls is a sensible base for Texas AI obligations, because the same questions recur: who or what accessed which data, what policy applied, and what the outcome was.

    For AI interactions, a useful evidence record captures the policy in force, the action taken (allowed, redacted, blocked or escalated), the identity of any human approver, and a timestamp. Retention periods and data residency should be configurable to match your obligations rather than fixed by a tool's defaults.

    Trussed enforces policy in the execution path across models, agents, tools and MCP servers, with actions that include allow, redact, block and escalate for human approval. Audit-ready evidence is generated by that enforcement and mapped to HIPAA, and deployment is self-managed in the customer cloud. It runs in production with healthcare organizations. This addresses the runtime group of obligations only. Disclosure, contracts and governance remain your own work.

    Practical next steps

    Start with an inventory of AI use cases across clinical, administrative and payer functions, including vendor-embedded features. Assign each to the law or laws that may reach it, and flag ambiguous cases for counsel. Then sort the resulting duties into runtime and non-runtime groups, assign owners, and confirm that each runtime duty has an enforcement point and an evidence record that an examiner could read without help from the engineering team.

    Re-check the statutory text and any agency or legislative developments on a fixed schedule. State AI law is still moving, and a mapping that was correct at one date may need revision.

    Map the use case first, then the statute

    TRAIGA (HB 149)

    General Texas AI act. Effective January 1, 2026.

    SB 1188

    Electronic health records, with provisions on practitioner use of AI in diagnosis and treatment.

    Your AI use cases

    Clinical decision support, diagnostic assistance, patient communications, utilization management.

    Your controls

    Runtime enforcement and evidence on one side, policy, disclosure and contracts on the other.

    Turn the mapping into enforceable controls and evidence

    Talk through which of your AI use cases need enforcement in the execution path and what audit-ready evidence your team would need to produce.

    Talk to an Expert