Statutory Foundations: How HB 1709 and Rule 165.1 Operate Together
Texas HB 1709 and TMB Rule 165.1 do not exist in isolation; they interlock to create a single documentation obligation for any Texas medical group that permits AI involvement in clinical notes. HB 1709, the Texas Responsible Artificial Intelligence Governance Act, governs how business entities—including healthcare organizations—deploy AI systems, and specialized guidance interpreting the statute establishes that any AI-modified clinical record must be clearly flagged so that subsequent treating physicians are not misled. TMB Rule 165.1, codified at 22 TAC §165.1, supplies the underlying medical record adequacy and retention standards against which that flagging is measured. Neither instrument treats AI labeling as an optional best practice; it is a statutory obligation enforced through the Board's record rules.
Under the combined framework, three obligations emerge for every AI-involved note. First, persistent AI labeling: each AI-authored or AI-modified entry must carry a clear indicator that travels with the record across all downstream formats for the full retention period. Second, non-misleading records: the flag must be visible at the point of care so any reviewing clinician understands where AI contributed text. Third, integration with the adequacy standard: the note must still be complete, contemporaneous, legible, and attributable to a physician observer, exactly as 165.1 has always required of paper and electronic charts alike.
Separate AI-care statutes further sharpen the physician's personal duty. Effective September 1, 2025, Texas law explicitly authorizes practitioners to use AI for care-related purposes, but only if they act within their license scope, comply with other applicable law, review all medical records created by AI to ensure conformity with TMB standards, and disclose their use of AI to patients. A companion provision effective January 1, 2026 requires clear and conspicuous disclosure of AI use in diagnostic or treatment-related services at the time care is first provided, subject to narrow emergency exceptions. For groups tracing these provisions against the broader literature on record integrity, the peer-reviewed corpus at https://www.ncbi.nlm.nih.gov/pmc/ offers useful grounding on why immutable amendment trails have long been the standard of defensible charting.
The Adequacy Standard: What Rule 165.1 Demands of Every Entry
Rule 165.1 defines an adequate medical record as complete, contemporaneous, and legible documentation of each patient encounter, maintained by each licensed physician regardless of medium. The rule enumerates required elements: the reason for the encounter and relevant history, physical examination findings and prior diagnostic results, the assessment or clinical impression, the plan of care including any discharge plan, and the date and legible identity of the observer. AI-generated content does not receive a lighter standard—if anything, it must clear the same bar while additionally satisfying the labeling and review overlays.
Electronic Record Requirements and the Amendment Rule
For electronic records specifically, Rule 165.1 imposes two provisions that are decisive for AI workflows. All non-biographical populated fields must contain accurate data based on the physician's actual findings and assessments—which forecloses any workflow that signs an AI draft without substantive review. And any amendment, supplementation, change, or correction not made contemporaneously must clearly show its time and date, indicate that it is an amendment or correction, and preserve the original text so that it remains legible. This amendment rule is the exact functional analog of a human-edit diff: the original AI draft is the base text, and every physician correction is a marked, timestamped, attributed change layered on top of it without erasing what came before.
Retention Timelines and Destruction Constraints
Retention under 165.1 runs a minimum of 7 years from the date of last treatment for adults. For minors, records must be kept until the patient reaches age 21 or 7 years from last treatment, whichever is longer. Critically, destruction is permitted only after any related civil, criminal, or administrative proceeding has been finally resolved. A group cannot purge on a fixed calendar schedule if litigation or a Board investigation is pending. Any AI documentation system whose logs and outputs are relied upon for clinical documentation effectively becomes part of the record ecosystem and inherits these same retention and destruction constraints.
Operationalizing Attestation: The Physician Review Workflow
Compliant Texas medical groups will generally adopt a workflow that converts the statutory review duty into recorded evidence. The AI scribe generates a draft note—HPI, ROS, physical exam, assessment and plan, orders, and counseling—but that draft is a pre-record artifact, not an official entry, until the physician reviews and approves it. The review interface should present a clear demarcation between AI-generated content and physician edits, whether through tracked changes, color highlighting, or a side-by-side comparison, and should expose the full version history from base draft through final signed note.
The attestation step is where the review duty becomes defensible. Before final sign-off, the physician completes a structured attestation—for example, an affirmation that the note has been reviewed, edited as necessary, and accurately reflects the clinical findings, assessment, and plan for the encounter. That attestation must record the physician's legible identity, the date and time, and a reference to the specific note version attested. Stored as immutable metadata bound to the note object rather than a generic system log, the attestation proves both that the physician discharged the AI-care review requirement and that the final record is a physician-approved document rather than an unvetted machine artifact. Groups can validate their attestation prompt language against statutory wording using the Scribing Template Directory before deployment.
Disclosure Propagation Across the Designated Record Set
HB 1709 requires a per-page disclosure that persists across the entire designated record set. When a physician signs an AI-derived note, the system should automatically tag the encounter resource as AI-involved and render a visible 'AI-Authored' label on every downstream page and format—printed charts, portal PDFs, faxed notes, C-CDA exports, and FHIR DocumentReference resources. If AI is later used to summarize or modify an existing note, each modified entry requires its own flag plus a traceable link to the original content, and the physician must re-review and re-attest to the modified version to remain within TMB standards.
Architecture Comparison: Manual, Generic, and Merry AI Approaches
The practical difference between documentation models becomes clear when the statutory obligations are placed side by side against how each approach actually behaves in production.
| Compliance Dimension | Manual Charting | Standard Generic AI Scribe | Merry AI Compliance Architecture |
|---|---|---|---|
| Per-page 'AI-Authored' disclosure (HB 1709) | Not applicable; no AI involvement | Often absent or buried in hidden metadata | Encounter-level flag renders visible header across print, PDF, fax, C-CDA, FHIR |
| Physician attestation object (AI-care statute) | Handwritten signature, no version binding | Generic click-to-sign, not bound to note version | Structured attestation with identity, timestamp, and immutable version reference |
| Amendment legibility (22 TAC §165.1) | Original often overwritten or illegible | Edits frequently overwrite the AI draft silently | Human-edit diff preserves original text; every change timestamped and attributed |
| Retention alignment (7-year floor) | Physical storage, purge-schedule risk | Fixed-calendar deletion common | Configurable WORM window anchored to last-treatment-date with litigation hold |
| Discovery readiness | Manual chart pull, gaps common | Logs not exportable in defensible form | Exportable version chain: draft, diffs, final, attestation for counsel and TMB |
The manual column carries no AI labeling burden but is prone to illegible amendments and calendar-based purge errors. The generic scribe column introduces AI without the statutory scaffolding—disclosures go missing, edits overwrite the draft, and audit logs are not structured for discovery. The Merry AI column treats each obligation as a discrete, testable behavior rather than a policy aspiration.
Software Requirements and the WORM Audit Trail
A defensible Merry AI configuration for Texas medical groups rests on a durable, write-once audit trail that captures the base AI draft, every physician edit expressed as a diff, the final signed version, and the associated attestation event. This design maps directly to the 165.1 amendment rule—the original text remains legible because the diff never destroys the base draft, and the identity and timestamp of each editor are recorded. There is also a deliberate separation between the transient inference layer, where raw audio and model context are shredded from RAM once the draft is produced and never written to disk, and the durable record layer, where the clinical text chain is preserved. The audio input is not part of the designated record set; the diff evidence is.
The one nuance groups must configure deliberately is retention duration. A 6-year WORM window alone is insufficient against the 7-year statutory floor, so Texas tenants must either extend the WORM period to match or exceed the last-treatment-date requirement or archive the trail into a secondary immutable repository preserving the same write-once characteristics for the full horizon, including any litigation hold. Overlay this technical stack with governance policies defining which encounters may use AI, mandatory review steps, prohibitions on signing without attestation, physician training on flag recognition, and periodic internal audits confirming every AI-involved note carries its flag, attestation, and diff history. When you are ready to validate your current configuration against these requirements, Book a 15-Minute Workflow Audit and we will trace a single encounter end-to-end against HB 1709 and Rule 165.1.
