Health systems are adopting ambient artificial intelligence (AI) scribes faster than patient consent practices have evolved to address how the technology records, processes, retains, and potentially reuses information from clinical encounters, according to a commentary published in The Lancet Digital Health. 

“The stance here is not against deployment of ambient scribes; rather, it is that deployment is outrunning the consent framework that would help patients to make an informed choice about participating in encounters in which ambient scribes are used,” wrote Mnotho Ngcobo of the University of Louisville Louis D. Brandeis School of Law, whose research includes the legal and ethical implications of AI in clinical decision-making. 

Consent gaps as adoption expands 

Ambient AI scribes record clinical conversations, transcribe and process them using a large language model, and convert the information into structured documentation stored in the electronic health record (EHR). Depending on vendor agreements, recordings also may be retained, analyzed for quality metrics, or used to train future models.  

Ngcobo wrote that health systems are increasingly integrating these tools into EHR platforms with minimal patient-facing disclosure, no standard opt-out process, and little transparency about data retention and secondary use. The commentary states that no uniform national standard governs patient disclosure, consent, data retention, or secondary use, and that professional societies have not established standards. 

Patient-centered outcomes also remain underexamined, according to Ngcobo. Published research has largely examined documentation burden, clinician burnout, and EHR efficiency, with comparatively few studies evaluating what patients understand about ambient scribes, how recording affects patient disclosure, or what information patients would want to know before agreeing.  

The commentary author acknowledged growing evidence that ambient scribes can reduce documentation burden. Ngcobo cited a multicenter quality-improvement study across six US health systems in which 30 days of AI scribe use was associated with reduced clinician burnout and improvements in cognitive task load and time spent documenting after hours. Other studies have reported reductions in note-writing time, after-hours EHR work, and documentation burden.

However, Ngcobo argued that efficiency gains for clinicians and transparency for patients are often treated as separate questions.  

Studies evaluating ambient scribes have used different approaches to patient disclosure, including verbal consent or notification that recording was taking place, according to Ngcobo. He noted that these approaches did not address whether patients understood who processed the audio, how long recordings were retained, or whether the information could be used beyond same-day clinical documentation.  

Patient awareness and trust 

Patient survey data cited in the commentary illustrate the gap. In a cross-sectional survey of 12,153 adults in Canada, 28% reported any awareness that AI scribes were being used during care, while 62% expressed reluctance toward their future use. Privacy concerns were among the strongest predictors of unfavorable attitudes involving comfort, trust, anticipated effects on patient-provider interactions, and willingness to use ambient scribes. 

The potential implications extend to what patients disclose during clinical encounters, according to the commentary. In two national surveys involving 4,510 US adults, 61% to 81% reported having withheld medically relevant information from a clinician at least once. Approximately one-third cited not wanting information disclosed during the encounter to be included in their medical record as a reason for nondisclosure. Ngcobo argued that introducing a recording that patients do not fully understand or might not have genuinely consented to could compound this dynamic. 

Attitudes also differ among patient populations. The commentary cited evidence that patients with lower income and education levels had less favorable attitudes toward AI scribes and that poorer general health was associated with lower odds of comfort and trust. “Thus, patients most in need of full and honest clinical encounters might be precisely those least likely to feel safe in recorded consultations,” Ngcobo wrote. 

Patients are not uniformly opposed to the technology. In the Canadian survey, 57% trusted AI documentation when human oversight was involved, and respondents with chronic conditions reported greater willingness to use ambient scribes. The commentary also noted a plausible equity argument that ambient scribes might reduce documentation disparities resulting from clinician bias or low language proficiency and could expand access to care in under-resourced settings where human scribes are unavailable. 

A framework for meaningful consent 

Ngcobo proposed three changes to address the consent gap across five areas: timing and content of disclosure, opt-out processes, secondary data use, and documentation of consent.

1. Patients should receive meaningful, plain-language disclosure before an ambient AI scribe is activated. The disclosure should explain what is recorded, who processes the information, how long it is retained, and what secondary uses are permitted, and it should be documented in the EHR. 

2. Patients should have a functional way to decline recording without penalty or awkwardness at the point of care. Their decision should be honored across subsequent visits without requiring them to repeat it. 

3. Secondary use of encounter data should be limited to direct clinical documentation unless patients separately and explicitly agree otherwise. Separate explicit consent should be required before recordings are used for purposes such as model training or billing analytics. 

Ambient AI scribes address a genuine problem, and the evidence that they help clinicians is credible and growing, Ngcobo wrote. He added that what remains unresolved is whether health systems can deploy technology that transforms clinical encounters into recorded and processable data without corresponding changes in how consent is obtained and understood. 

“Patients who speak most candidly in examination rooms are those who believe that what they disclose will go no further than necessary, and tools designed to support clinical care should not quietly undermine the trust on which it depends,” he wrote. 

Ngcobo declared no competing interests.