AI Governance Staff Training

Enter your information to begin.

Estimated time: 45–60 minutes including knowledge check

1

What AI Does in Your Practice

Your practice uses AI tools that listen to patient visits and generate clinical notes. The AI creates a DRAFT — not a finished record.

Your provider reviews it, corrects errors, and signs it. Only then does it become a legal medical record.

Your role depends on your position, but everyone has a part in making sure AI is used safely and correctly.

Key point: AI generates a draft. Your provider's review and signature are what make it a medical record.

2

What Can Go Wrong

AI can fabricate clinical findings that never happened. Published studies show error rates above 25% across major platforms. Every error in a signed note is a potential liability.

Fabricated Findings

Exam details you never performed documented as if they happened.

Invented Medications

Drugs never prescribed listed in the patient's record.

False History

Past medical history or allergies that don't exist.

Wrong Codes

Billing codes inflated beyond what the visit supports.

Every hallucination you miss = potential malpractice liability and patient harm.

3

Your Signature = Your Liability

When a provider signs an AI-generated note, it carries the same legal weight as a handwritten note.

The AI vendor is not liable — their contracts explicitly disclaim accuracy. The provider who signed it is liable.

Insurance carriers are now adding AI exclusion endorsements. If your policy has one and an AI-related claim is filed, the claim gets denied. You bear the full cost.

Governance documentation is what carriers, attorneys, and regulators evaluate. This training is part of that documentation.

4

Provider Responsibilities

THE DEPOSITION TEST: If this note was presented in court as evidence against you, could you defend every sentence?

Review every note word-for-word — not a skim
Verify all exam findings — did you actually perform them?
Verify all medications — did you actually prescribe them?
Document your verification (amendment, addendum, or signature note)
Never sign a note you can't fully defend

5

Staff Responsibilities

Medical Assistant

AI notes are drafts with errors. Flag obvious problems before the provider signs. Make sure patient disclosure is documented. Know how to disable the AI system in an emergency.

Front Desk

Give every patient the AI disclosure notice. Document their consent or opt-out. If they opt out, flag the record immediately.

Billing

Never submit AI-suggested codes without provider verification. AI frequently recommends inflated codes. $28,619 per false claim under the False Claims Act.

6

The Verification Standard

NOT ENOUGH ✗

✗ Skim the note and sign

✗ Trust AI to be accurate

✗ Assume EMR captured everything

✗ Sign without reading

✗ Rely on vendor assurance

DEFENSIBLE ✓

✓ Read every sentence carefully

✓ Verify findings you performed

✓ Check completeness against your exam

✓ Amend if you find errors

✓ Document your verification process

Can you defend every word in court?

7

Patient Disclosure and Opt-Out

Every patient must be informed AI is used in their care.

1

Notice

Provide AI disclosure

2

Consent

Patient decides

3

Document

Record in EHR

4

Review

Provider checks

5

Proceed

AI or manual

If a patient opts out: Document immediately. Flag the record. Switch to manual documentation. Notify all staff. Patient choice is legally binding.

8

When Something Goes Wrong

Error Correction Protocol:

1

STOP

Don't sign

2

IDENTIFY

Note the error

3

CORRECT

Fix it

4

DOCUMENT

Record the fix

5

SIGN

Corrected version

6

REPORT

If pattern

Kill Switch — disable AI immediately when:

Patient harm event

Systematic errors detected

Data breach suspected

Regulatory inquiry received

9

Incident Reporting

REPORT IMMEDIATELY

Patient harm or adverse outcome

Critical AI system failure

Suspected data breach

WITHIN 24 HOURS

Caught and corrected error

Note hallucination or fabrication

Systematic AI inaccuracy

Include in every report: date, time, which AI system, what the error was, what impact it had, what you did about it.

10

Why This Protects You

Following these protocols isn't just about the practice — it protects you personally.

Malpractice claims: Documented governance shows you followed proper procedures
Insurance reviews: Governance documentation demonstrates responsible AI use
Regulatory audits: Your compliance record is already built
Personal liability: Following protocols gives you individual evidence

Certification under AI Governance Shield™ means your practice's governance has been independently validated. This training is part of that certification. Annual renewal keeps it current as laws and AI tools evolve.

Questions about your practice's certification? Contact us at sentinelriskgrp.com

11

Behavioral Health & Substance Use — Extra Protections

If your practice provides any behavioral health, psychiatric, or substance use disorder (SUD) treatment — or if you see patients via telehealth who receive those services — two additional layers of regulation apply.

42 CFR Part 2 — Federal Confidentiality of SUD Records

Substance use disorder treatment records have stricter federal protections than standard HIPAA. Key points every staff member must know:

Separate consent required. A general HIPAA authorization does NOT cover SUD records. Patients must give specific, written consent before SUD information can be disclosed — even to other providers.
AI systems don't know the difference. An AI scribe will document substance use history, treatment details, and medication-assisted therapy the same way it documents anything else. If that note is shared without proper Part 2 consent, the practice is in violation.
Re-disclosure prohibition. Anyone who receives Part 2 information cannot re-disclose it. If an AI-generated note containing SUD information is sent to a referring provider or insurer without consent, every downstream disclosure is also a violation.
Practical rule: Before signing any note involving SUD treatment, verify the patient's Part 2 consent is on file and current. Flag SUD records in your EHR so they are not shared through health information exchanges without proper consent.

Tennessee SB 1580 — First Private Right of Action for Mental Health AI

Effective July 1, 2026, Tennessee law creates the first statute allowing patients to sue directly over AI misuse in behavioral health:

$5,000 Per Violation

Each improper use of AI in behavioral health treatment is a separate violation.

Treble Damages

Willful violations = triple damages. No annual cap.

No Independent Decisions

AI cannot make therapeutic decisions without direct provider oversight and approval.

Telehealth Reach

Applies if the patient is in Tennessee, regardless of where the provider is located.

What this means for you: If you treat any patient in Tennessee via telehealth — or if your practice provides behavioral health services anywhere — AI-generated therapy notes, treatment plans, and psychiatric evaluations require the same word-for-word verification as medical notes, plus you must confirm the AI did not generate independent clinical recommendations. The provider makes every therapeutic decision. The AI only documents what the provider decided.

12

State AI Disclosure Laws — Texas TRAIGA (and the pattern)

Texas adopted the first comprehensive state AI law for providers. If your practice treats any patient located in Texas — including by telehealth — this applies to you.

What TRAIGA requires: a healthcare provider that uses an AI system in relation to a patient's care or treatment must disclose that use to the patient (or their representative) no later than the first date the AI-assisted service or treatment is provided. In emergencies: disclose as soon as reasonably possible.

What enforcement looks like: the Texas Attorney General enforces it. After a 60-day notice-and-cure window, civil penalties can reach $200,000 per violation that cannot be cured — and ongoing violations can accrue daily penalties.

Why this matters to YOUR role: the practice's AI disclosure notice — the one front desk delivers and documents — is not just good practice. In Texas it is the law, with a clock attached. Every state is watching this pattern: disclosure duties tied to first use, with state AG enforcement. Your practice's disclosure workflow is built to satisfy the strictest state that applies to your patients.

Telehealth rule of thumb (Section 8 of the check): the patient's state decides which law applies — a Texas patient on a video visit brings TRAIGA with them.

13

AI Bias Recognition — What to Watch For

AI tools can produce systematically different output for different patient groups — even when no one intended discrimination. The federal nondiscrimination rule under § 1557 (operative July 5, 2024) requires every covered practice to make reasonable efforts to identify and mitigate this risk. Recognizing the patterns is the workforce's part.

Your role: Notice patterns. Flag them to the Designated AI Officer. You are not expected to run statistics. Your eyes on the chart are the practice's first line of detection.

Patterns to watch for in AI output:

Pronoun & Gender Errors

AI consistently mis-gendering specific patient groups, defaulting to one gender, or flattening gender identity in notes.

Cultural / Language Drop-off

AI documenting non-English-speaking patients or LEP patients with notably less detail or skipping cultural context that's clinically relevant.

Demographic Coding Patterns

AI suggesting systematically lower-acuity codes for one racial group, or systematically higher codes for another — when the clinical picture is comparable.

Pain & Symptom Bias

AI documenting reported pain, distress, or psychiatric symptoms differently across demographic groups — a known systematic issue in clinical AI.

Referral / Triage Disparities

AI scheduling or triage tools systematically slowing access for one patient group, or routing them to different care paths.

Disability Accommodation

AI failing to capture accommodations, communication needs, or accessibility-relevant context for patients with disabilities.

What is NOT bias to flag: A single error in a single note is just an error — that's the verification standard from earlier sections. Bias is a systematic pattern — same kind of error happening repeatedly with the same patient demographic. Patterns over a week or month are what to flag.

Reporting pathway when you observe a pattern:

1

DOCUMENT

Note the pattern + sample evidence

2

ESCALATE

To Designated AI Officer same week

3

SCRUTINIZE

Apply elevated review until resolved

4

FOLLOW UP

Confirm vendor / fix landed

The practice runs a quarterly fairness review (Policy 13 § V) where the Designated AI Officer formally examines AI output for these patterns. Your in-the-moment observations feed that review. Together, that's how the practice meets its § 1557 affirmative-effort obligation — consistent with federal AI governance guidance (NIST AI RMF MEASURE function).

Knowledge Check — 14 Questions

You need 12 out of 14 correct (~86%) to pass.