AI Tools for US Doctors in 2026: HIPAA-Compliant Stack for Physicians
AI helps US physicians with ambient scribing, clinical documentation, patient communication, prior authorization, and medical literature review. The non-negotiable: any AI touching PHI must operate under a signed BAA. HIPAA-compliant tools include Abridge, Nuance DAX Copilot, Suki, Doximity GPT, and OpenEvidence. Verified May 2026.
Hard floor for US physicians
Never input Protected Health Information into any AI tool without a signed Business Associate Agreement. Consumer ChatGPT, Claude, and Gemini do not come with a BAA. Not medical advice. AI does not replace clinical judgment, board certification, or the standard of care.
GPTPrompts.AI Editorial
GPT Prompts editorial team. Not medical, legal, or compliance advice. Confirm BAA status and your institution's AI policy before deploying. · Last updated May 23, 2026
What HIPAA requires before you use AI
The HIPAA Privacy and Security Rules govern any use of Protected Health Information (PHI) by US physicians, practices, and health systems. Any third party that creates, receives, maintains, or transmits PHI on behalf of a covered entity is a Business Associate and must sign a Business Associate Agreement (BAA) before PHI is shared. This is the legal floor for every AI tool that touches a patient encounter, message, chart, or identifier.
- HIPAA Privacy and Security Rules: Define what counts as PHI and what protections you owe patients.
- Business Associate Agreement (BAA): Required in writing before any AI vendor handles PHI on your behalf.
- 21st Century Cures Act and information blocking: Affects how AI-generated content interacts with patient access to records.
- FDA AI/ML guidance (SaMD): Diagnostic AI is regulated as Software as a Medical Device, a different category from admin tools.
- State medical board rules on AI: Disclosure, supervision, and scope of AI use varies by state; check your board.
- CMS reimbursement: Some AI-augmented services have specific CPT and HCPCS codes; rules change year to year.
- ONC certification: EHR-integrated AI features should run inside an ONC-certified workflow for safety and auditability.
This is general information, not legal advice. Your privacy officer, general counsel, and malpractice carrier are the right people to confirm the specifics for your practice.
10 AI tools US physicians actually use
BAA status, EHR integration, and price for each. Verified May 23, 2026 against vendor documentation. Pricing changes often; confirm with the vendor before purchase.
| Tool | Best use | HIPAA / BAA | EHR integration | Cost |
|---|---|---|---|---|
| Abridge | Ambient clinical scribe (visit to note) | Yes, BAA standard for enterprise deployments | Epic native (App Orchard), Cerner/Oracle integration | Enterprise pricing, contact sales |
| Nuance DAX Copilot | Ambient scribe inside Microsoft and Epic workflows | Yes, BAA via Microsoft/Nuance | Epic native, Cerner/Oracle, Meditech | About 600 dollars per clinician per month |
| Suki | Voice assistant plus ambient scribe | Yes, BAA standard | Epic, Cerner/Oracle, Athena, Meditech, eClinicalWorks | About 399 dollars per clinician per month |
| DeepScribe | Ambient scribe with specialty templates | Yes, BAA standard | Epic, Cerner/Oracle, Athena | Custom per practice |
| Doximity GPT | Drafts notes, letters, and patient handouts (no PHI) | BAA-aware product design, but never paste PHI | No direct EHR integration (drafting only) | Free for verified US physicians |
| OpenEvidence | Clinical Q&A grounded in peer-reviewed literature | Not a PHI tool, no BAA needed for literature search | No EHR integration | Free for verified US clinicians |
| ChatGPT Plus | General drafting and study only, never PHI | No BAA on consumer Plus | None | 20 dollars per month |
| ChatGPT Enterprise | General enterprise AI with BAA option | BAA available via Enterprise agreement | API integrations possible, not EHR native | Custom pricing (typically 60 to 75 dollars per user per month) |
| Microsoft 365 Copilot | Email, document, and meeting AI inside M365 | BAA via Microsoft enterprise agreement | Not EHR native, used for office workflows | 30 dollars per user per month (enterprise terms) |
| Google Vertex (Med-PaLM 2 and successors) | Developer platform for building clinical AI | BAA via Google Cloud Healthcare API | Requires custom integration and dev work | Usage-based, plus integration cost |
FDA-cleared diagnostic AI is a different category. This page covers admin and documentation tools, not diagnostic Software as a Medical Device.
The 8 AI workflows US physicians actually deploy in 2026
Each workflow below assumes you have a signed Business Associate Agreement with the vendor before any Protected Health Information is involved. Where the workflow can run on de-identified or generic content, that is called out explicitly. Not medical advice. AI does not replace clinical judgment, board certification, or the standard of care.
1. Ambient scribing during patient visits
The single highest-yield US physician workflow in 2026. The clinician opens an app (Abridge, DAX Copilot, Suki, or DeepScribe) at the start of the visit, the AI listens to the natural conversation between physician and patient, then produces a structured note ready for review and EHR sign-off. Selection criteria: BAA in place, EHR integration that matches your shop (Epic vs Cerner/Oracle vs Athena), specialty template quality, latency, and per-clinician cost. Pilot with two willing physicians for four weeks before a full rollout. Never deploy a scribe that lacks a signed BAA.
2. Clinical note drafting (SOAP, H&P)
Even without ambient capture, AI inside a BAA-covered tool can draft a SOAP note or History and Physical from a short structured input. The physician reviews, edits, and signs. This is appropriate inside Abridge, DAX Copilot, Suki, ChatGPT Enterprise (with BAA), or Microsoft 365 Copilot (with BAA). It is not appropriate inside consumer ChatGPT, Claude, or Gemini because there is no BAA on those products. The signed clinician is still the legal author of the note and remains responsible for accuracy.
3. Patient communication drafts (portal messages)
Portal message volume is the number-one driver of physician burnout in many practices. AI integrated into your EHR (Epic’s In Basket AI is the most common example in 2026) drafts a reply that the clinician reviews and edits before sending. Outside the EHR, draft non-PHI templated language in Doximity GPT or a BAA-covered tool. Never paste a patient message containing PHI into a consumer AI tool. Disclose AI assistance per your institution’s policy and applicable state law.
4. Prior authorization letter generation
Prior auth is one of the highest-ROI uses of AI for US physicians in 2026 because the work is templated, repetitive, and well-documented. Tools that sit inside the EHR or that you feed a de-identified clinical summary (and that have a BAA when PHI is involved) can produce a first draft of the medical necessity letter, including relevant CPT and ICD-10 codes and cited guidelines. The physician reviews and signs. Do not let AI fabricate citations: every guideline reference must be checked.
5. Medical literature review
OpenEvidence is the default in 2026 because it is free for verified US clinicians, grounded in peer-reviewed medical literature, and includes inline citations to journal articles. UpToDate and DynaMed remain the institutional standards. Consumer ChatGPT can summarize public literature, but it hallucinates citations and must never see PHI. Use it for background reading on a public topic, then verify every citation in PubMed before relying on it. Cite the original article, not the AI summary.
6. Patient education handouts (plain-language summaries)
AI is well-suited to translating a clinical concept into 6th-grade reading level handouts in English and Spanish. Generic, non-PHI handouts can be drafted in any AI tool. Patient-specific handouts that reference the patient’s diagnosis, labs, or medication list must use a BAA-covered product. Always include a line directing the patient back to their physician for personal medical questions, and add the standard not-medical-advice disclaimer per your institution’s policy.
7. Coding and billing query support (ICD-10, CPT)
AI can suggest ICD-10 and CPT codes from a de-identified clinical summary and explain the difference between similar codes. Treat the output as a draft that a human coder verifies. CMS audits and OIG enforcement focus on documentation supporting the code, so the underlying note (not the AI suggestion) is the record. Never up-code based on AI output you have not verified. Compliance officers should review any AI-assisted coding workflow before it goes live.
8. CME study tools and board-prep
ChatGPT Plus, Claude Pro, Doximity GPT, and OpenEvidence are all used for board review and CME study. Because no PHI is involved in study, the BAA requirement does not apply (you are using public study material). Useful uses include explaining a guideline change, working through a clinical vignette, or quizzing yourself on differential diagnoses. Always cross-check against your CME source of truth, because AI can confidently produce incorrect medical facts.
The 3 mistakes US physicians make with AI
- Pasting PHI into consumer ChatGPT, Claude, or Gemini. No BAA, no exceptions. Even a single chart entry is a HIPAA violation. Use a BAA-covered enterprise product or stick to de-identified content.
- Signing AI-drafted notes without reading line by line. The signing clinician is the legal author. AI hallucinates findings, drug doses, and citations more often than physicians notice on a fast scan.
- Treating an unregulated chatbot as clinical decision support. FDA-cleared CDS is a regulated category (SaMD). General purpose chatbots are not cleared for diagnosis. Use them for study and background, not for binding clinical decisions on a real patient.
When AI is NOT appropriate
- Any task involving PHI in a tool without a signed BAA.
- Diagnosis from an unregulated chatbot (no FDA clearance, no SaMD review).
- Drug dosing or prescribing decisions without verifying in a primary reference (Lexicomp, Micromedex, FDA label).
- Medical coding submitted to payers without human coder review and documentation that supports the code.
- Communications with patients in a high-acuity situation where the conversation needs your judgment, not a draft.
- Anywhere your state medical board or institution prohibits AI use, or where patient consent is required and absent.
- Any clinical situation where you would not be comfortable defending the decision in a malpractice action.
The verdict: what I actually run for US physicians in 2026
For most US physicians I work with in 2026, the right stack is one ambient scribe with a signed Business Associate Agreement (I usually recommend piloting Abridge for Epic shops, DAX Copilot for Microsoft-heavy systems, and Suki when price matters), Doximity GPT for free non-PHI drafting, and OpenEvidence for free clinical literature search. I keep ChatGPT Plus around at 20 dollars per month for study and templates, but I never enter PHI into it because there is no BAA. If you remember one rule from this page, remember this: never input PHI into any AI tool without a signed BAA. The rest is workflow tuning. Verified May 23, 2026.
AI for US physicians FAQ
What are the best AI tools for US doctors in 2026?
For ambient scribing, the top three picks are Abridge, Nuance DAX Copilot, and Suki. For literature search, OpenEvidence stands out: it costs nothing for verified US clinicians and stays grounded in peer-reviewed sources. For general drafting that involves no PHI, Doximity GPT comes free to credentialed US physicians. For enterprise AI under a signed BAA, ChatGPT Enterprise, Microsoft 365 Copilot, and Google Vertex on Google Cloud Healthcare API are the usual picks. Verified May 2026.
Is ChatGPT HIPAA compliant for physician use?
The consumer tiers of ChatGPT (the Free, Plus, and Pro plans) are not HIPAA compliant because OpenAI will not sign a Business Associate Agreement for those products. The Enterprise tier is the version that offers a BAA. Without it, you must never enter Protected Health Information into the product, even de-identified data that could be re-identified. Use the consumer tier strictly for study, generic templates, and non-PHI tasks. If your workflow needs ChatGPT for PHI, your organization needs the Enterprise contract.
Abridge vs DAX Copilot vs Suki: which ambient scribe should I pick?
All three are HIPAA-compliant under a signed BAA and ship integrations with the major EHRs. Abridge has the deepest Epic native integration and strong specialty coverage. Nuance DAX Copilot is the default for Microsoft-heavy shops, priced near 600 dollars monthly per clinician, and benefits from Microsoft and Epic distribution. Suki sits at roughly 399 dollars monthly per clinician as the price-friendly option, and it supports more EHRs. The right answer depends on your EHR, specialty, and budget. Run a four-week pilot with two willing clinicians first.
What is Doximity GPT and is it safe for clinical use?
Doximity GPT is a free AI tool inside the Doximity app for verified US physicians, intended for drafting notes, letters, patient handouts, and appeals. It is designed with physician privacy in mind, but the product itself is not an EHR and the recommended posture is de-identified drafting only. Never paste a patient name, MRN, date of birth, or other PHI into the tool. Use it for templates, generic patient education content, study material, and administrative writing where no PHI is required.
Can I use ChatGPT to write patient notes?
Only if you are working entirely with de-identified information, or you are inside the Enterprise tier covered under a signed BAA. On the consumer tiers there is no BAA, so entering Protected Health Information would breach HIPAA. The safer pattern: use a purpose-built scribe (Abridge, DAX, or Suki) for the actual chart note, and keep consumer ChatGPT to templates, study, and non-PHI drafting. The signing clinician owns the chart entry as the legal author and cannot delegate that liability to software.
Which AI tools help with prior authorization in 2026?
Prior auth is one of the highest-yield AI workflows for US physicians because the work is templated and well-documented. Products that sit inside an EHR under BAA, or BAA-covered standalone vendors fed only a de-identified summary, can draft the medical necessity letter, attach relevant guideline citations, and suggest appropriate CPT and ICD-10 codes. Always double-check every cited guideline yourself, because models sometimes invent references. The treating physician reviews and signs the final letter as the responsible clinician on record.
What about AI for clinical decision support and FDA rules?
Clinical decision support that meets the definition of a medical device falls under FDA regulation as Software as a Medical Device (SaMD). The 21st Century Cures Act carves out some CDS that surfaces guidelines a clinician can independently verify, but diagnostic AI typically requires FDA clearance or approval. The admin and documentation tools covered on this page are not diagnostic devices. Never rely on an unregulated chatbot for diagnosis. Confirm SaMD status with the vendor and your compliance team.
How much does AI cost for a solo physician vs a group practice?
A solo doctor can run a useful stack for under 50 dollars total monthly: free Doximity GPT, free OpenEvidence, plus either ChatGPT Plus (20 dollars monthly, no PHI) or Claude Pro (20 dollars monthly) for non-PHI work. Adding an ambient scribe pushes the total to roughly 400 to 700 dollars per clinician each month. Group practices typically negotiate volume pricing on the scribe vendor of choice, and they pair it with an enterprise productivity suite covered under their own signed BAA.
What are the EHR integration realities for AI scribes?
Epic is the most common in large US health systems, and Abridge plus DAX Copilot have the deepest Epic native integrations in 2026. Oracle Cerner is the second most common; Suki, Abridge, and DAX all support it. Athena, eClinicalWorks, and Meditech are also supported by Suki and several others. Standalone scribes that copy and paste into the EHR work but lose efficiency. Confirm your EHR vendor, version, and whether the integration is App Orchard certified before purchase.
What is the malpractice exposure for AI-assisted documentation?
Whoever signs remains responsible for everything in the chart, the prescription, and any prior auth letter. AI assistance does not transfer responsibility. Errors that lead to patient harm remain attributable to the physician under negligence law. Carriers in 2026 generally accept AI-assisted documentation when the clinician reviews and signs, when a BAA is in place, and when the institution has a written AI policy on file. Confirm coverage and policy with your malpractice carrier before deploying any clinical AI workflow.
What is the best AI for medical literature search in 2026?
OpenEvidence is the standout choice for American clinicians: at no cost to verified physicians, anchored in peer-reviewed journal content, and inline citations point straight to the source articles. UpToDate and DynaMed remain the institutional standards for point-of-care reference work. Consumer ChatGPT can summarize public literature for background reading, but it invents citations more often than clinicians realize, so every reference must be checked in PubMed before being cited. Never paste any patient-identifying detail into a literature search tool.
When should I NOT use AI in clinical practice?
Skip AI for any task where the safety floor is not met: pasting PHI into a tool without a Business Associate Agreement, relying on a non-FDA-cleared chatbot for a diagnosis, using AI as the source of truth for drug dosing without verifying in a primary reference, signing AI-drafted notes you have not read line by line, and acting on AI coding suggestions without human coder review. Also skip AI when your state medical board or institutional policy expressly prohibits the use, because AI policy varies across US states.
Related healthcare AI guides
HIPAA-aware AI for the nursing workflow
Bedside-safe ways nurses can use ChatGPT
Front-desk and back-office AI for clinics
The full healthcare AI landscape overview
Mental health prompt library with privacy notes
HIPAA-compliant AI for mental health clinicians
Keep reading
Related guides you'll like
- Industry Guides
AI Tools for Dental Offices US
AI workflows for dental front desk scripts, patient education drafts, recall messages, reviews, and privacy-safe admin
Read guide → - Industry Guides
AI Tools for Accountants US
US guide to AI tools for accountants, bookkeepers, and CPA firms covering client emails, checklists, SOPs, and tax-season workflows
Read guide → - Industry Guides
Best AI Tools for Consultants US
Buying guide for consultant AI tools across discovery, research, proposals, slides, and delivery operations
Read guide → - Industry Guides
AI Tools for Nonprofit Directors
US guide to AI tools for nonprofit directors covering fundraising, grants, board reports, volunteers, and donor privacy
Read guide → - Industry Guides
AI Tools for Medical Billing US
AI workflows for denial summaries, patient billing explanations, payer call prep, appeals, and privacy-safe RCM work
Read guide → - Data & Research
AI Tools for Competitive Intelligence
AI workflows for competitor monitoring, battlecards, market scans, and source-backed briefs
Read guide →