AI Informed Consent in Emergency Medicine: A Practical Script for ED Physicians

Medically Reviewed by Chester Shermer, MD, FACEP | Published May 21, 2026 | 9 min read

AI informed consent in emergency medicine is no longer optional. As of 2026, both the AMA’s ethical framework and a growing set of state laws, including California’s AB 3030 and Texas’s TRAIGA and SB 1188, require physicians to disclose AI use in diagnosis or treatment to patients before or at the time of interaction. The harder question is not whether to disclose, but how to do it in 60 seconds at the bedside in language a patient in pain can actually process. This guide gives you a practical script and the legal context behind it.

Why AI Consent Is Different Now

For decades, decision-support tools such as automatic EKG readers, rule-based risk calculators, and sepsis alerts have shaped emergency care without explicit patient consent. The legal and ethical environment has changed, and three forces are converging.

First, the regulatory floor moved. Beginning January 1, 2025, California’s AB 3030 required disclosure when generative AI is used to communicate clinical information to patients.[1] Effective in late 2025 and early 2026, Texas SB 1188 and TRAIGA require physicians to disclose AI use in diagnosis or treatment, with emergency situations specifically requiring disclosure as soon as reasonably practicable after the interaction.[2]

Second, professional ethics caught up. The AMA’s principles on augmented intelligence state that when AI directly impacts patient care, access to care, or medical decision-making, that use should be disclosed and documented to physicians and patients in a culturally and linguistically appropriate manner.[3]

Third, patient trust is fragile. A 2026 CHAI/NORC national survey found that while most Americans already use AI in some form, only a small minority feel very comfortable with it in healthcare. Many report that AI involvement makes them trust healthcare less, and the overwhelming majority report at least one concern about AI in clinical care.[6]

The implication is straightforward: patients are increasingly likely to ask whether AI was involved in their care, and the legal framework increasingly requires that you tell them whether they ask or not.

What Has to Be Disclosed, and What Does Not

Not every algorithm requires explicit consent. The emerging consensus, including a JAMA-published framework on AI disclosure, draws the line based on whether the patient has meaningful agency to respond to the disclosure and the degree of physical risk involved.[5]

Disclosure is generally required when:

• AI is used in diagnosis or treatment decisions that the physician acts on.
• AI is used to communicate clinical information directly to the patient.
• AI is involved in tools the patient interacts with directly.
• An ambient AI scribe is recording the encounter.

Disclosure is less commonly required when:

• AI is embedded in longstanding decision-support tools such as automated EKG interpretation or rule-based risk scores, although guidance is moving toward broader disclosure.
• The communication is reviewed by a licensed clinician before reaching the patient, which is an explicit exemption under California AB 3030.[1]

The defensible approach is simple: when in doubt, disclose. The downside of over-disclosure is a 30-second conversation. The downside of under-disclosure is a complaint to the medical board, malpractice exposure, and a rapidly deteriorating physician-patient relationship.

The Seven Domains You Should Understand Before Disclosing

A peer-reviewed framework specific to emergency medicine identified seven areas an emergency physician should understand about any AI tool used in care in order to secure meaningful informed consent.[4]

• How the AI system actually operates.
• Whether it is understandable and trustworthy.
• The known limitations and error modes.
• How disagreements between physician and AI get resolved.
• Whether the patient’s identifiable information and the AI system are secure.
• Whether the system has been validated.
• Whether the system exhibits known bias.

You will not cover all seven in a 60-second emergency department disclosure. You should still know them before you stand at the bedside, because the patient who asks a substantive question deserves a substantive answer.

For more on bias specifically, see Algorithmic Bias in ED Triage: What Every Emergency Physician Needs to Know.

A Working ED Consent Script

The challenge in the emergency department is time. The patient is in pain, frightened, or both. You may have only 30 to 60 seconds. The script below is designed to be modular so you can use only the portions that apply to the AI tool actually in play.

Opening

“Before we get started, I want to let you know that we use some AI-assisted tools in this department to help with documentation and to support clinical decisions. I review everything the AI suggests, and the final medical decisions are mine. Do you have any questions about that?”

If using an ambient AI scribe

“While we talk, I’m going to have an AI tool listening in to help me with the chart. It helps me focus on you instead of on a keyboard. The recording is encrypted, it does not leave our system, and I review everything before it goes in your record. If you would prefer I not use it, just say the word.”

If AI is being used in a specific diagnostic decision

“For your [chest pain, stroke symptoms, etc.], we run the imaging and labs through an AI tool that helps flag findings I might otherwise miss. I review the results myself. If we disagree, my clinical judgment is what decides your care.”

If a patient asks, “Is AI making my decisions?”

“No. AI is one input, like a lab result or a colleague’s opinion. I make the decision. I can override the AI any time my judgment says to, and I do that when the patient in front of me does not match what the AI is seeing.”

If a patient declines

“That’s fine. We have a workflow that does not use this tool. It might take a little longer for me to chart, but it will not change the care you receive.”

For a bedside framework on when physician judgment should supersede machine output, see When to Override the AI: A Decision Framework for Emergency Physicians.

What to Document

A verbal disclosure matters only if the chart reflects it. The simplest pattern includes three pieces.

• A standing dot-phrase or smart-phrase documenting: “AI-assisted [scribe/decision support/imaging interpretation] was used in this encounter. Patient was informed and given an opportunity to decline. Patient [consented/declined]. All AI output reviewed and verified by the treating physician.”
• A standing departmental protocol that defines which AI tools are in use, what disclosure script applies, and how declines are handled.
• A governance record that lists the tools in active use and the date the consent workflow was last reviewed.

For more on governance structure, see Building an AI Governance Framework for Your Emergency Department.

Common ED Scenarios

Trauma activation or unstable patient

You will not obtain pre-encounter consent on an intubated trauma patient. Both the AMA framework and TRAIGA accommodate this. TRAIGA explicitly states that emergency disclosure must occur as soon as reasonably practicable after the event.[2] Document that the patient could not be consented at presentation because of acuity, then disclose to the patient or surrogate once the clinical situation permits.

Pediatric patient with a parent present

Disclose to the parent or legal guardian. For older children and adolescents, consider including them in the conversation using age-appropriate language. Document who received the disclosure.

Limited English proficiency

CHAI’s 2026 patient survey emphasized that disclosure paired with clear explanation builds trust, while disclosure without explanation can erode it.[6] Use a qualified medical interpreter. A two-sentence disclosure delivered clearly through an interpreter is far stronger, legally and ethically, than a longer disclosure the patient does not fully understand.

Patient with capacity questions

If the patient lacks decision-making capacity, disclose to the legally authorized representative. If no representative is available and care is emergent, proceed under the standard emergency exception and document disclosure as soon as the patient regains capacity or a representative is reached.

Common Mistakes to Avoid

• Treating disclosure as a one-time form. AI tools change, and your consent workflow has to reflect what is actually in use today.
• Outsourcing disclosure to registration staff. Disclosure of AI in diagnosis and treatment is a physician-patient conversation, not a waiting-room sign.
• Disclosing only when the patient asks. Current ethical and legal guidance frames disclosure as physician-initiated.[2] [3]
• Overpromising accuracy. Avoid saying that the AI is “very accurate.” A better statement is that AI is one tool you use and that you review its output.
• Failing to document. From a medico-legal standpoint, a consent conversation that is not in the chart effectively did not happen.

Dr. Chet’s Take

The hardest part of AI informed consent in the emergency department is not the legal piece. It is the time piece. You have a patient with chest pain, the room next door has a febrile two-year-old, and there is a hallway bed waiting for a triage upgrade. Adding even 30 seconds of structured disclosure to every encounter sounds easy until you actually try to do it repeatedly during a busy shift.

My advice is to invest the work once. Build a department-standard script. Train the team to use it. Embed it in your dot-phrases. Make declining easy and friction-free. Once the workflow is in place, the disclosure costs 15 seconds, not 60, and you have a defensible, ethical, and patient-centered process that will stand up to regulatory review, board scrutiny, and malpractice review alike.

The bigger trust problem in medicine right now is not whether we use AI. It is whether patients believe we are honest with them about it. That part you can fix this week.

— Chester Shermer, MD, FACEP | Emergency Medicine, 25+ Years Clinical Experience

Key Takeaways

• AI informed consent in the emergency department is increasingly mandated by both state law and professional ethics.
• Disclosure should generally occur when AI affects diagnosis, treatment, communication with the patient, or tools the patient interacts with directly.
• A practical disclosure script can be delivered in 30 to 60 seconds if the workflow is standardized.
• Trust improves when disclosure is paired with explanation, accountability, and visible physician oversight.
• A consent conversation that is not documented in the chart is unlikely to protect you later.

Frequently Asked Questions

Do I legally have to tell patients when I use AI in their care?

It depends on the state and the specific use case. As of 2026, California requires disclosure when generative AI communicates clinical information to patients, and Texas requires disclosure when AI is used in diagnosis or treatment.[1] [2] The AMA’s ethical framework recommends disclosure whenever AI directly impacts patient care, regardless of state law.[3]

What specifically should I disclose?

At minimum, disclose that an AI tool was used, what category it falls into, that the physician reviews the output and retains decision authority, and that the patient may decline when appropriate. TRAIGA also emphasizes that disclosure should be conspicuous and in writing where practicable.[2]

How do I handle an unconscious or unstable patient?

You do not obtain traditional pre-encounter consent. Emergency disclosure exceptions permit disclosure as soon as reasonably practicable after the clinical situation stabilizes. Document why pre-encounter consent was not possible, then disclose to the patient or surrogate once feasible.[2]

Does an ambient AI scribe require its own consent process?

In practice, yes. State recording-consent laws, HIPAA expectations, and guidance from medical groups and vendors all support a specific disclosure about how the conversation is processed, whether audio is retained, and the patient’s option to decline.[8]

What if my hospital does not yet have an AI consent policy?

Raise it through your AI governance committee or, if none exists, through department leadership and risk management. In the meantime, document your own disclosures using a standardized phrase. The absence of a policy does not eliminate the disclosure obligation.

Will telling patients about AI scare them away from care?

Available data suggests the opposite when disclosure is paired with explanation. Patients who are told nothing and later discover AI involvement tend to react more negatively than patients who were informed at the time of care.[6]

Medical Disclaimer

This content is intended for licensed medical professionals, EMS personnel, and trained emergency responders. It is not legal advice. Specific consent requirements vary by state, by institution, and by the AI tool in use. Consult your institutional legal counsel, compliance officer, and AI governance committee for jurisdiction-specific guidance.

Continue Your Training

Ready to put this into practice?

If you are an emergency physician trying to understand how AI will actually impact your clinical practice, not just the hype, explore the training resources built for real-world deployment and physician-led oversight.

Browse All Courses at Global MedOps Command

Relevant Reading for This Topic

• How to Avoid Becoming an AI Casualty
• Emergency Department Efficiency Playbook
• AI Literacy in Emergency Medicine: What Every EM Physician Needs to Know
• Algorithmic Bias in ED Triage: What Every Emergency Physician Needs to Know
• When to Override the AI: A Decision Framework for Emergency Physicians
• How to Evaluate an AI Tool Before Using It in Your Emergency Department
• How to Implement an AI Ambient Scribe in Your Emergency Department
• AI Change Management in the Emergency Department
• Physician Leadership Through the AI Transition

Connect With Dr. Shermer

LinkedIn – Chester Shermer, MD, FACEP

References

1. Medical Board of California. GenAI Notification Requirements (AB 3030).
2. Akerman LLP. New Year, New AI Rules: Healthcare AI Laws Now in Effect.
3. Infectious Disease Advisor. AMA Issues New Principles for Use of AI in Medicine.
4. Iserson KV. Informed consent for artificial intelligence in emergency medicine. PubMed.
5. AI Disclosure and Patient Consent in Health Care. JAMA Network.
6. Coalition for Health AI. CHAI Releases New Patient Survey Report on Health AI and Transparency.
7. Stanford HAI. Ethical Obligations to Inform Patients About Use of AI Tools.
8. MGMA. Patient Consent Form for AI Dictations.