Low-Intensity Focused Ultrasound for Opioid Use Disorder: Early Evidence and Open Questions.

Ambient artificial intelligence (AI) “scribes” systems that record clinician‐patient conversations and automatically generate clinical notes, are being deployed at unprecedented speed in US health systems []. Tools such as Nuance DAX Copilot, Abridge, and other generative AI‐based documentation assistants have expanded rapidly across large academic centers, integrated delivery networks, and multisite physician groups [, ]. In one large integrated practice group more than 2.5 million patient‐encounter notes were generated within a year following deployment []. Early multicenter evaluations consistently show that ambient AI documentation reduces after‐hours charting, improves perceived documentation‐related well‐being, and may reduce clinician burnout at scale [, ]. In an environment where documentation burden is framed as an existential threat to care quality, retention, and financial stability, AI scribe adoption appears not merely attractive but inevitable. Yet despite this intense commercial and institutional momentum, far less attention has been paid to the systemic and infrastructural implications of outsourcing the first mile of clinical documentation (the moment when the clinical encounter becomes clinical data) to proprietary AI systems. This matters because the first mile of documentation is where clinical reality is translated into institutional knowledge. Algorithmic mediation at this stage can introduce systematic biases, standardizations, or omissions that propagate across registries, dashboards, and models, effectively steering learning health systems along paths that reflect infrastructural constraints rather than clinical truth. Most published studies focus on frontline outcomes such as workflow satisfaction, burnout reduction, or self‐reported efficiency []. These are important but incomplete indicators of successful adoption. What remains largely unexamined is how ambient AI scribes reshape what gets written into the electronic health record (EHR), how it is written, and by whom. These questions matter because EHR documentation is not only a legal and billing artifact; it is the epistemic substrate of learning health systems (LHSs) []. Errors introduced at this layer therefore carry direct patient safety implications, not only for downstream analytics but for the clinicians and patients who depend on accurate documentation to guide care. Emerging evidence raises concerns that AI‐generated notes may introduce hallucinated clinical details, omit safety‐critical information, or differentially misrepresent encounters with certain patient groups, especially individuals with diverse accents, nonstandard speech patterns, or limited English proficiency [, , ].