AI Healthcare Compliance: The rules governing AI at the point of care
An AI scribe drafts the note, but the physician signs it. We track the consent rules, malpractice exposure, and payer policies that decide what a clinician owes before and after the tool runs.
A topic page in AI Regulation News, our source-backed tracker of the rules that change how professionals use AI at work.
AI reached the exam room faster than almost anywhere else in professional work, and the rules are catching up in a way that lands directly on the physician. Ambient scribes now draft the clinical note while the visit happens, AI tools summarize records and suggest orders, and payers run automated systems that decide prior authorization. None of that moves the core duty. The note carries the physician's signature, the diagnosis is the physician's call, and the liability for what the record says stays with the clinician who attests to it. The compliance question is not whether to use these tools. It is what the clinician owes before the tool runs and after it produces a draft.
Two patterns define this lane. The first is consent and privacy. An ambient scribe that records a visit is capturing a conversation, and in states with two-party consent laws that recording can trigger wiretap exposure if the patient was not properly informed. The protected nature of health information raises the stakes further. The second pattern is the review window. An AI-drafted note is a draft, not a record, until the physician reviews and signs it, and a rushed or rubber-stamp attestation is where malpractice risk concentrates. An error the model introduced becomes the physician's error the moment it is signed without a real read. Plaintiffs and boards are beginning to treat the quality of that review as the question.
The third front is the payer side, where AI shifts from a tool the clinician uses to a system the clinician fights. Automated prior-authorization and coverage decisions, including under Medicare pilots, can deny care at machine speed, and the practical burden falls on the practice to appeal. Understanding how those systems decide, and building a faster evidence-backed appeal, is now part of running a compliant and financially viable practice.
The workflow consequence is a short, non-negotiable routine around every AI-assisted encounter: confirm and document patient consent before an ambient scribe records, treat every AI-drafted note as a draft that requires a genuine review against what actually happened in the visit, attest only after that read, and keep a defensible appeal process ready for AI-driven denials. Wire consent and review into the visit flow and the speed is a gain rather than a liability. The briefings below track the specific consent rulings, malpractice signals, and payer policies setting these duties. This is clinical workflow interpretation, not legal or medical advice, and it is the healthcare lens distinct from the broader tracker.
The briefings we track in this lane
California AB 1979, SB 503: Clinical AI Duties
Signed Sept 30, 2026: once operative, AB 1979 would protect clinician judgment over clinical AI tools and SB 503 would set reasonable-effort bias duties.
Read the briefing AnalysisAI Note Malpractice and the Review Window Before You Sign
An AI-drafted note is a draft until you attest to it. Where malpractice risk concentrates, and the review step that keeps the model's error from becoming yours.
Read the briefing LitigationAI Scribe Wiretap Lawsuits and What Physicians Owe Now
Ambient scribes record the visit, and two-party consent states can turn that recording into wiretap exposure. The consent step to run before the tool listens.
Read the briefing Agency policyMedicare WISeR and Winning AI Prior-Auth Appeals Faster
Automated prior-authorization can deny care at machine speed. How the WISeR model decides and how to build a faster, evidence-backed appeal.
Read the briefingBrowse the full tracker for every rule, ruling, and agency action we follow across professions.
Keep going
- Ukraine Ministry Draws AI Red Lines for PsychologistsNothing in law.
- Swiss DPAs Issue AI Care-Monitoring GuidanceTwo supervisory bodies that between them cover private and public care institutions have put a written position on AI-assisted patient monitoring on the record: local, in-room processing as the default, cloud o
- BfDI Asks Berlin to Legislate Health Data AIA federal supervisory authority put a request for AI-specific health data legislation on the record, about seven weeks before an opt-out research extraction from the electronic patient record is stated to begin
- FDA discussion paper on generative AI-enabled medical devicesNon-binding. Comments close 19 October 2026 under docket FDA-2026-N-7874 on a competency-based premarket concept.
- Read what AI scribe wiretap lawsuits mean for physicians
- Browse the full AI Regulation News tracker
Recent healthcare AI coverage
- Washington board: only humans may practice medicineWashington's Medical Commission opened consultation on two drafts: only humans may hold a medical license, and AI misuse may risk discipline.
- Virginia HB 481: Physician Must Approve Prior-Auth DenialsVirginia HB 481 bars insurers from denying prior authorization unless a licensed physician reviews it, blocking AI-only denials.
- Vietnam Drafts Rules for AI Use of Medical DataVietnam's health ministry is drafting rules that would gate secondary use of patient data for AI. What vendors should prepare for.
- Vermont Act 156: AI Mental Health LimitsVermont Act 156 bars AI from independently providing mental health services. AI-only therapy is unprofessional conduct.
- Texas AG Settles AI Health-Claims CaseThe Texas AG settled with Pieces Technologies over deceptive accuracy claims for a hospital AI tool. The first state-AG AI enforcement.
- CMS Proposes Software as a Medical Service Pay RuleCMS's proposed rule creates a Software as a Medical Service payment class for AI clinical software. Comments close August 2026.
- MHRA: an AI scribe becomes a device on the claimThe MHRA set out when an AI scribe counts as a medical device in Great Britain. The line falls at intended purpose and marketing claims, not technology.
- UK MHRA Expands Its AI Medical-Device SandboxThe MHRA funded its AI Airlock sandbox to build the pathway for AI as a medical device, testing risk classification and bias metrics.
- Texas TRAIGA HB 149: AI Disclosure Duty in HealthcareTexas TRAIGA (HB 149) requires disclosing AI use in clinical care. What providers must tell patients, the emergency exception, and the documentation step.
- Texas SB 1188 Requires Physicians to Disclose Diagnostic AITexas SB 1188, effective Sept 1 2025, lets practitioners use diagnostic AI in scope but must review AI-created records and disclose AI use to patients.
- Texas SB 815 Bars Automated AI Utilization DenialsTexas SB 815 bars automated AI from making utilization-review denials without a human reviewer. What the law requires of payers and what it means for care.
- Tennessee SB 1580: AI Cannot Pose as a TherapistTennessee SB 1580 makes it a deceptive act for AI to claim it is a qualified mental health professional. $5,000 per violation.
- Taiwan's TFDA Rules for AI Diagnostic DevicesTaiwan's TFDA guideline for AI and machine-learning detection and diagnosis devices changes what hospitals and vendors must validate before clinical use.
- Korea's First Approval Path for Generative Medical AISouth Korea published what its regulator calls the world's first approval pathway for generative-AI medical devices. What makers must submit.
- Singapore Updates Health AI RulesSingapore's MOH and HSA published AIHGle 2.0, assigning accountability, transparency and risk duties to healthcare AI developers and users.
- Rhode Island AI Clinical Note Disclosure LawRhode Island H7538 now requires providers to tell patients when AI documents a visit and to review the note for accuracy. In force June 22, 2026.
- Rhode Island Bars AI From Therapy Decisions: 2026 LawTwo Rhode Island laws make AI chatbots flag suicide risk and bar AI from therapeutic decisions without a licensed clinician.
- NZ Medical Council: Doctors Own AI in Patient CareThe Medical Council of New Zealand says doctors may use AI only within their competence and stay responsible for clinical decisions.
- Nevada AB 406: AI Barred From Direct Mental Health CareNevada AB 406 bars AI from providing mental or behavioral health care directly to patients and bans AI therapist claims. What clinicians must do now.
- Maine LD 2082 Limits AI in Mental Health CareMaine LD 2082 restricts AI in therapeutic communication and adds consent rules for mental health care. What is prohibited and when it takes effect.
- Japan's IDATEN Rules for Self-Updating Medical AIJapan approves self-updating medical AI only under IDATEN, with a pre-agreed change plan and post-market monitoring.
- Indonesia Is Drafting Oversight Rules for Hospital AIIndonesia's health ministry is preparing a Good AI Medical Practice regime that would validate hospital algorithms before use.
- Indiana HB 1271: AI Downcoding Now Needs a Human ReviewIndiana's HB 1271 bars insurers from using AI as the sole basis to downcode a claim without a human reading the record, and requires disclosure.
- India's AI Medical Software Rulebook Keeps the Filing StepCDSCO's final Medical Device Software guidance formalises an Algorithm Change Protocol for AI, but model updates still need approval or notification.
- Georgia to Bar AI-Only Health Insurance Denials in 2027Georgia SB 444 bars AI from issuing a health-coverage denial without a clinical peer's review. Signed, effective January 1, 2027.
- FDA Guidance: When Clinical AI Is a Regulated Medical DeviceFDA's revised guidance sets a four-criteria test for when clinical AI is a regulated medical device. What physicians must check before they deploy.
- EU MDCG Guidance on AI Medical Devices and the AI ActMDCG 2025-6 explains how the EU AI Act, MDR and IVDR apply together to AI medical devices. Non-binding guidance.
- Delaware Bars AI From Holding a Medical License or TitleDelaware HB 191, signed April 23 2026, bars a nonhuman entity or AI agent from being licensed as a physician, PA, or nurse, or using those titles.
- Colorado Extends Insurer AI Bias Rules to Auto and HealthColorado's Amended Regulation 10-1-1 extends insurer AI governance and antidiscrimination duties to auto and health plans from July 1, 2026.
- CMS MIPS Now Credits Tracking AI Patient-Safety EventsA new CMS MIPS improvement activity credits clinicians for tracking AI-attributable patient-safety events. What IA_PSPA_34 asks and how to document it.
- Health Canada's Rules for Machine-Learning DevicesHealth Canada makes machine-learning device makers declare AI use and file a predetermined change control plan.
- Ontario's Medical College: Consent Before AI RecordingOntario's College of Physicians requires documented patient consent before ambient AI recording, and holds the physician accountable.
- California SB 1120 Puts Physicians Over AI in Coverage CallsCalifornia SB 1120, effective Jan 1 2025, requires a licensed clinician, not AI, to make medical-necessity denials in health-plan utilization review.
- California AB 3030 Requires AI Disclaimer on Patient MessagesCalifornia AB 3030, effective Jan 1 2025, requires a prominent AI disclaimer and a way to reach a human when GenAI writes clinical patient messages.
- Brazil's CFM Makes Doctors Liable for AI Care Aug 26Brazil's Federal Council of Medicine makes physicians responsible for AI-assisted decisions and requires recording AI use.
- TGA: Full SaMD Rules Apply to AI Medical SoftwareAustralia's TGA says AI features do not create a new device class. AI medical software still owes full Software as a Medical Device duties.
- Arizona Bars AI-Only Medical Necessity DenialsArizona HB 2175 makes an insurer medical director review medical-necessity denials individually, curbing AI-only claim denials.
- Alabama SB63: Licensed Clinician Required on AI Coverage DenialsAlabama SB63 takes effect October 1, 2026. A licensed clinician must make every AI-assisted denial, plus annual certification and disclosure duties.
- Medicare Asks If AI Breaks Its 1992 Payment BasisIn the CY 2027 Physician Fee Schedule CMS says generative and agentic AI were not anticipated when RVUs were set, and seeks comment on revaluation.
- CMS Asks Whether AI Data Firms Need a CLIA LicenceCMS and CDC CLIA request for information asks whether data-only facilities need a certificate and how labs verify AI tools. Comments 14 September.
- Nursing Boards Adopt Model AI Rules That Bind NobodyNCSBN adopted Model Act revisions on 21 August 2026. The AI rule governs nursing education programmes, and binds nobody until a state adopts it.