As ambient AI tools increasingly move from pilot programs into everyday clinical workflows, health systems are beginning to confront a complex question: When AI-generated documentation is wrong, who is responsible?
Chief medical information officers across several health systems said the answer is not simple, but most agree the clinician remains the final checkpoint before information enters the medical record.
Ambient AI documentation tools listen to patient visits and generate draft clinical notes, aiming to reduce administrative burden and free clinicians from typing during appointments. But the technology can also produce inaccurate summaries, omit key details or introduce errors that clinicians must identify before signing the note.
For many organizations, that reality has shaped how they define accountability.
“At VHC, we are guided by a fundamental principle: Ambient AI output is considered a draft,” said Usman Akhtar, MD, CMIO at Arlington, Va.-based Virginia Hospital Center. “The clinician maintains accountability for any documentation that is signed and subsequently becomes part of the medical record.”
Dr. Akhtar said his organization has built safeguards into the deployment of these tools, including mandatory review protocols before signing notes and reporting mechanisms that allow clinicians to flag issues. If patterns of errors emerge, they are addressed through clinical and IT governance processes and escalated to vendors for remediation.
Several CMIOs said the issue of documentation accuracy is not entirely new in healthcare.
J. Clay Callison Jr., MD, vice president and CMIO at Knoxville-based University of Tennessee Medical Center, noted that EHRs have long introduced their own forms of documentation risk. Earlier examples include errors carried forward in problem lists or transcription mistakes from speech-recognition software — sometimes referred to informally as “dragonisms.”
“With generative AI, the challenge could become much larger and trickier,” Dr. Callison said. “Generative AI content may be more disguised than simple phonetic errors.”
Even so, he said responsibility still lies with the clinician caring for the patient because they are the only person who fully understands the clinical situation.
“Ensuring that the clinician receives the feedback will be critical as we move forward,” he said.
While clinicians remain the final gatekeepers, health systems said they also carry responsibility for how these tools are implemented and monitored.
Amer Saati, MD, CMIO of Roseville, Calif.-based Adventist Health, said his organization approaches accountability through a layered governance model.
“The clinician remains ultimately responsible for attesting to the accuracy of the documentation,” he said. “But the health system owns oversight of implementation, monitoring and safe deployment, including performance review, audit processes and corrective workflows.”
If errors appear systemic — suggesting problems with the underlying model rather than isolated mistakes — the vendor becomes part of the accountability chain.
“That becomes a shared accountability issue with the vendor,” Dr. Saati said, addressed through feedback loops, contractual expectations and ongoing performance monitoring.
Other CMIOs described a similar three-part structure: Clinicians verify the final note, health systems manage deployment and governance and vendors are responsible for improving the underlying technology.
“Clinicians are ultimately accountable for what enters the medical record, but they must be supported by accurate, reliable tools,” said Priti Patel, MD, CMIO at Walnut Creek, Calif.-based John Muir Health. “Vendors are responsible for building high-quality tools and continuously monitoring performance, while health systems are responsible for thoughtful vendor selection and training clinicians on appropriate use.”
When errors occur, Dr. Patel said organizations are increasingly relying on formal feedback channels that allow clinicians to flag problems so vendors can correct them at the source.
As ambient AI becomes more deeply embedded in healthcare operations, CMIOs said these governance structures will become increasingly important. The technology may automate documentation, but it does not eliminate the need for human oversight.
Instead, many leaders said the future of AI documentation will depend on clearly defined roles: clinicians verifying the record, health systems managing safe deployment and vendors continuously improving the tools that power the system.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.