TX Rule 165.1 AI Scribe Disclosure: 2026 Compliance Playbook
Comply with Texas Medical Board Rule 165.1 by authenticating every AI-drafted note and disclosing AI use at first service; this audit confirms a defensible 2026 workflow.

Texas Medical Board Rule 165.1 AI Scribe Medical Record Disclosure Requirements: 2026 Operations Playbook
Texas Medical Board Rule 165.1 (22 TAC §165.1) requires every AI-drafted note to be complete, contemporaneous, legible, and authenticated by the treating physician. Under SB 1188 and the Texas Responsible AI Governance Act, clinics must disclose AI use to patients no later than first service and retain records seven years.
This playbook is written for Clinic Medical Directors who own documentation policy, Board exposure, and payer audit risk across a Texas practice. It turns the medical records rule, the 2025 legislative session's AI statutes, and current federal coding policy into one operating procedure your physicians, scribes, and compliance staff can follow on Monday morning.
Rule 165.1 makes the signing physician accountable for every AI-generated sentence
CLINICAL UPDATE 2026: Revised for 2026 CMS CPT G2211 attestation rules, California SB 1120 compliance, and NCCI Modifier 25 bundling criteria.
The Texas Medical Board does not regulate AI scribes as a separate category of tool. It regulates the record. Under 22 TAC §165.1(a), each licensed physician must maintain an adequate medical record for each patient that is complete, contemporaneous, and legible. The rule makes no exception for text produced by software, so the Board will evaluate an ambient AI draft exactly as it would evaluate a note you typed yourself once you sign it.
An adequate encounter record under Rule 165.1 should document the reason for the encounter and relevant history, physical examination findings, and prior diagnostic results; an assessment, clinical impression, or diagnosis; a plan for care; and the date and legible identity of the observer. The rule also expects documentation of patient progress, response to treatment, relevant risk factors, medication frequency and dosage, referrals, patient education, and follow-up instructions.
The "legible identity of the observer" element is where AI scribes create the most quiet risk. An ambient model is not an observer. The physician who examined the patient is the observer, and the note must make that unmistakable. If a draft describes a finding the physician never elicited (a hallucinated negative Lachman test, for example), the signature converts that fabrication into the physician's sworn clinical record.
The rule further expects that CPT and ICD-10-CM codes reported on claims be supported by the documentation in the record. This links Board compliance directly to billing integrity. An AI note that inflates history or examination elements to support a higher level of service is both a records violation and a potential false claim.
Retention obligations do not change because a draft originated in software. Rule 165.1(b) requires physicians to retain records for at least seven years from the date of last treatment, and for minors until age 21 or seven years from last treatment, whichever is longer. Medical directors should decide in writing whether the unedited AI draft, the audio, and the transcript are part of the designated record set or are transient working files destroyed after signature.
Texas now layers three statutes on top of Rule 165.1
Senate Bill 1188, enacted in 2025, added provisions to the Health and Safety Code governing electronic health records and artificial intelligence. It permits a practitioner to use AI for diagnostic purposes, including recommendations on diagnosis or treatment based on a patient's record, when the practitioner acts within the scope of licensure, the use is not otherwise prohibited, the practitioner reviews all AI-created records consistent with Texas Medical Board medical records standards, and the practitioner discloses the use of AI to patients.
House Bill 149, the Texas Responsible Artificial Intelligence Governance Act (TRAIGA), took effect January 1, 2026. It requires a health care service provider to disclose to a patient, or the patient's personal representative, that an AI system is being used in relation to a health care service or treatment. The disclosure must be clear, conspicuous, written in plain language, and free of manipulative design. It must be made no later than the date the service is first provided, or as soon as reasonably possible in an emergency.
Texas Penal Code §16.02 establishes one-party consent for recording conversations. Because the physician is a party to the encounter, recording itself is generally lawful in Texas without patient consent. That is a wiretap answer, not a disclosure answer. TRAIGA and SB 1188 impose transparency duties that operate independently of wiretap law, so "Texas is one-party consent" is no longer a complete compliance position.
Federal HIPAA requirements still apply in full. Your AI scribe vendor is a business associate, the audio contains protected health information, and a signed business associate agreement is mandatory before the first encounter is captured. For a jurisdiction-by-jurisdiction view of consent, recording, and disclosure rules, use the State AI Scribe Consent & Compliance Vault.
Why ambient documentation should trigger disclosure
Some counsel argue that ambient scribes perform documentation rather than diagnosis, placing them outside SB 1188's diagnostic language. TRAIGA's disclosure duty is broader, covering AI used "in relation to" a health care service. Modern scribes also suggest diagnoses, problem lists, and codes. The defensible 2026 posture is to disclose every ambient AI use and document that disclosure.
Patient disclosure must occur at first service and appear in the chart
A disclosure that lives only in a lobby poster is difficult to prove during a Board complaint investigation. The operational standard should be a written, plain-language notice delivered at intake (paper or portal), plus a brief verbal confirmation by the physician at the start of the recorded encounter, with a timestamped entry in the record.
Effective disclosure language should state four facts: that an AI tool listens to and drafts documentation of the visit, that the physician reviews and is responsible for the final note, how long audio is retained (or that it is deleted after note generation), and that the patient may decline without any effect on care. Avoid phrasing that buries the AI reference inside general privacy language, since TRAIGA requires clarity and conspicuousness.
Opt-out handling needs a defined workflow. When a patient declines, the medical assistant should flag the encounter, the ambient device should be disabled, and the physician should document manually or with a non-AI human scribe. The opt-out should persist across future visits until the patient changes it, and the chart should show the preference.
Intake notice signed or acknowledged: stored as a scanned document or portal attestation.
Verbal confirmation at encounter start: captured in the transcript and summarized in the note.
Structured disclosure field populated: date, time, method, and patient response.
Representative disclosure for minors: delivered to the parent or legal guardian.
Emergency exception documented: disclosure completed as soon as reasonably possible afterward.
Physician review attestation converts an AI draft into a compliant record
SB 1188 explicitly conditions lawful AI use on the practitioner reviewing all AI-created records consistent with Texas Medical Board standards. Rule 165.1 is that standard. A signature applied in bulk at the end of clinic, without reading the note, fails both requirements and is the single most common failure pattern in ambient scribe deployments.
A defensible review attestation reads something like: "I personally performed the history and examination documented above. This note was drafted with an AI documentation tool and was reviewed, edited, and verified by me for accuracy and completeness." Place it above the signature line so it is visible to auditors, not hidden in metadata.
The contemporaneous requirement deserves attention. Rule 165.1 does not specify an hour-count deadline, but notes signed days or weeks after the encounter invite challenge. Set an internal policy (for example, signature within 24 to 72 hours) and monitor compliance by physician. AI scribes generally shorten time to signature, which strengthens your contemporaneous position when the policy is enforced.
Edits must be traceable without becoming late alterations. Changes made before signature are ordinary authorship. Changes made after signature must follow your addendum procedure, with date, time, author, and reason. Never overwrite a signed note to correct an AI error. Board investigators treat undocumented post-signature changes as record tampering.
High-risk elements every physician should verify
Examination findings never performed: AI tools sometimes infer normal findings from templates.
Medication doses and frequencies: Rule 165.1 expressly expects these specified.
Laterality and anatomic site: critical for procedure coding and surgical history.
Attribution of statements: family member versus patient statements must be distinguished.
Diagnoses suggested but not confirmed: remove or label as differential.
Audit exposure concentrates in seven documentation parameters
The table below maps each Rule 165.1 and 2026 statutory parameter to the specific documentation evidence required, the vulnerability auditors and Board investigators exploit, and the structural mechanism Scribing.io uses to close the gap at the document object model (DOM) level of the note.
Texas AI scribe compliance parameters mapped to audit risk | |||
Clinical Parameter | Documentation Requirement | Audit Vulnerability | Scribing.io DOM Mechanism |
|---|---|---|---|
Patient AI disclosure (TRAIGA, SB 1188) | Clear, plain-language notice no later than first service; patient response recorded | No proof disclosure occurred; generic privacy notice deemed not conspicuous | Required disclosure node with timestamp, method, and response; note cannot finalize while empty |
Observer identity (Rule 165.1(a)) | Date and legible identity of the examining physician on each encounter | AI authorship ambiguous; human scribe or tool mistaken for observer | Author and observer fields separated; AI tool logged as drafting agent only |
Physician review (SB 1188) | Evidence the practitioner reviewed all AI-created content | Batch signing; identical notes across patients; sign time under seconds | Section-level review markers and dwell logging before signature unlocks |
Examination findings | Findings actually elicited, with laterality and pertinent negatives | Cloned or templated normal exams contradicting the transcript | Each finding linked to its transcript source; unsupported sentences flagged amber |
Code support (Rule 165.1, NCCI) | CPT and ICD-10-CM codes supported by documented work | Modifier 25 without separately identifiable E/M; G2211 without longitudinal plan | Separate procedure and E/M blocks; G2211 relationship statement prompt |
Post-signature changes | Dated, attributed addenda with reason for change | Silent overwrites interpreted as alteration | Signed DOM locked; edits generate versioned addendum objects only |
Retention and residency (Rule 165.1(b), SB 1188) | Seven-year retention; EHR data physically stored in the United States | Offshore processing or storage; audio retained without policy | U.S.-only processing and storage; configurable audio deletion with audit log |
Coding accuracy depends on physician-verified AI output, not transcription
Rule 165.1's expectation that codes be supported by documentation means every AI coding suggestion is a liability until the physician confirms it. Three 2026 billing areas generate the most AI-related denials and recoupments in Texas clinics: Modifier 25 on same-day procedures, the G2211 visit complexity add-on, and diagnosis specificity.
Modifier 25 requires a separately identifiable E/M
Consider an established Medicare patient with right knee primary osteoarthritis, M17.11 (ICD-10-CM), who receives a corticosteroid injection (CPT 20610) and also has E11.9 (ICD-10-CM), type 2 diabetes without complications. Under NCCI policy, the procedure includes the pre-procedure decision to inject and routine assessment of the knee. Billing an E/M with Modifier 25 requires significant, separately identifiable work.
An AI scribe that blends the knee discussion and the diabetes management into one narrative undermines the claim. The defensible note separates them: one block documents the knee evaluation and injection (consent, site, laterality, medication, dose, lot, tolerance), and a distinct block documents diabetes management, such as A1c review, medication adjustment, and the effect of corticosteroid injection on glycemic control. The E/M level should then be supported by the diabetes work alone.
G2211 needs a documented longitudinal relationship
HCPCS G2211 recognizes the inherent complexity of visits that are part of an ongoing longitudinal relationship serving as the continuing focal point for a patient's care, or ongoing care for a single serious or complex condition. CMS has expanded payment for G2211 when the base E/M is reported with Modifier 25 for certain preventive services, but the add-on still requires the relationship to be evident in the record.
Ambient tools rarely capture this relationship unprompted, because physicians do not say it aloud. Medical directors should require a short relationship statement in the assessment and plan, such as "I serve as the continuing focal point for this patient's diabetes and musculoskeletal care and will manage longitudinally." Without it, G2211 claims are vulnerable on post-payment review.
Diagnosis specificity must match the documented findings
AI models sometimes over-specify or under-specify codes. Coding M17.11 requires the note to establish unilateral, primary, right knee osteoarthritis. Coding E11.9 asserts the absence of documented complications; if the transcript mentions neuropathy or nephropathy, E11.9 is inaccurate and the more specific E11 subcategory applies. Physicians should treat suggested codes as drafts subject to the same review as narrative text.
SB 1188 requires Texas EHR data stored in the United States
Beginning January 1, 2026, SB 1188 requires covered entities to ensure that electronic health record data for Texas patients is physically maintained in the United States or a U.S. territory, including data stored by third-party or cloud service providers. This reaches beyond the EHR itself to vendors that process and store clinical content on your behalf.
Ambient AI pipelines involve several data stages: audio capture, transcription, model inference, draft storage, and EHR transfer. Each stage is a potential offshore exposure if the vendor uses global cloud regions or international subprocessors. Ask every vendor for a written data-flow map, subprocessor list, and contractual U.S.-residency commitment covering all stages, then attach it to the business associate agreement.
Audio retention policy should be explicit. Many clinics delete audio once the note is signed, which reduces breach exposure and storage cost. Others retain audio for quality review. Either choice is defensible if written, applied consistently, and reflected in the patient disclosure. Undefined retention is the indefensible option.
A 30-day implementation sequence brings a Texas clinic into compliance
The following sequence assumes an ambient AI scribe is already deployed or about to launch. It prioritizes the controls that reduce Board and payer exposure fastest, then layers in training and monitoring.
Days 1 to 5, inventory: list every AI tool touching clinical content, including scribes, coding assistants, and portal chatbots.
Days 1 to 5, contracts: confirm business associate agreements and U.S. data residency language for each vendor.
Days 6 to 10, disclosure: finalize plain-language intake notice, verbal script, and opt-out workflow; load into intake and EHR.
Days 6 to 10, attestation: standardize the physician review statement and place it above the signature block in every template.
Days 11 to 15, templates: rebuild specialty templates to separate procedures from E/M and to prompt G2211 relationship statements; the 1,000+ Specialty Clinical Macros library offers starting points.
Days 16 to 20, training: train physicians on high-risk verification elements and the addendum procedure; train staff on disclosure and opt-out.
Days 21 to 25, policy: adopt a written AI documentation policy covering signature timeliness, audio retention, and designated record set scope.
Days 26 to 30, audit: sample ten signed AI notes per physician against the transcript; score accuracy, disclosure, and code support.
Ongoing monitoring should be quarterly. Track disclosure completion rate, median time to signature, frequency of post-signature addenda, AI-flagged unsupported sentences left unedited, and Modifier 25 and G2211 denial rates. A physician whose notes show near-zero edits across hundreds of encounters warrants a focused review, since it often signals unread signatures rather than perfect AI output.
Multi-state groups must reconcile Texas disclosure with stricter consent states
Texas permits one-party recording, but a group with telehealth patients or clinics in other states inherits the strictest applicable rule for each encounter. A Texas physician conducting a video visit with a patient located in an all-party consent state should obtain affirmative consent before recording, not merely provide disclosure.
Pennsylvania illustrates the difference clearly. Its wiretap statute generally requires consent of all parties to intercept an oral communication, and violations carry criminal and civil exposure. Review Pennsylvania Wiretapping Statutes for Ambient Scribes before enabling ambient capture for any Pennsylvania-located patient.
California adds payer-side AI rules. SB 1120 (the Physicians Make Decisions Act) requires that health plan and insurer medical necessity determinations be made by a licensed physician or qualified health care professional, not by an AI tool alone. Texas groups contracting with California plans should expect AI-generated documentation to be scrutinized by human reviewers, raising the stakes on note accuracy.
The practical solution is a single enterprise standard set at the strictest common denominator: affirmative consent captured in the record, plain-language disclosure at intake, physician review attestation, and U.S. data residency. Applying one workflow everywhere is simpler to train and easier to defend than state-by-state variation.
Medical directors should treat the AI scribe as a supervised delegate
The governing principle across Rule 165.1, SB 1188, and TRAIGA is consistent: the physician remains the author of record, the patient has a right to know AI is involved, and the record must reflect what actually happened in the exam room. An AI scribe is a drafting delegate under physician supervision, much like a human scribe, and should be governed with the same discipline.
Clinics that structure the note itself around these requirements (disclosure fields that cannot be skipped, findings linked to source audio, attestation above the signature, locked records with versioned addenda) convert compliance from a training burden into a default. That design approach is the basis of the Scribing.io Clinical Interface, built so Texas physicians sign notes that satisfy the Board, the statute, and the payer at once.
Review this playbook at least annually and after each Texas legislative session, since the 2025 session showed how quickly AI obligations can change. Confirm current rule text with the Texas Medical Board and qualified health care counsel before finalizing policy, and document each review so your compliance program itself shows the contemporaneous diligence the rules expect.

