AI in Therapy: Convenience vs. Privacy
The Incident That Raised the Alarm
In Fayetteville, Arkansas, a librarian named Molly Quinn reported that her therapist began using an AI note-taking tool during a session in November. The tool recorded the session without Quinn's explicit consent—a practice she had previously expressed discomfort with.
The recording was done without my explicit consent, and that broke my trust.
As a result, Quinn ended the therapeutic relationship entirely.
How the Tools Work
Several companies—including Berries, SimplePractice, and Blueprint—now offer AI systems that record therapy sessions, generate transcripts, and draft clinical notes. Berries CEO Tal Salman emphasized privacy safeguards: audio is automatically deleted after each session, transcripts are stored on HIPAA-compliant servers in the U.S., and therapy content is not used to train AI models.
The Privacy Paradox: HIPAA Is Not Enough
Kellie Owens, assistant professor of medical ethics at NYU Grossman School of Medicine, warned that HIPAA compliance does not eliminate data breach risks. She noted that written consent alone may not ensure an informed choice, and that verbal discussion about recording is essential.
Written consent can be a box-checking exercise. Patients need a real conversation about what they're agreeing to.
Therapist Concerns: Trust and Accuracy
Marisa Cohen, a couples and sex therapist in New York, raised two key issues:
- The presence of AI may alter what clients disclose, making them less open or honest.
- Errors in AI-generated notes could become part of a permanent medical record, with no easy way to correct them.
The Efficiency Trade-Off: Time Saved vs. Oversight Needed
Kym Tolson, a therapist practicing remotely, reported that AI documentation reduced her administrative time from 15–20 minutes per client to about two minutes. However, she stressed that she still reviews and edits every AI-generated note, and that she discusses the recording process with clients beforehand.
Public Trust in AI for Mental Health Care Is Low
Recent surveys highlight widespread skepticism:
- A YouGov survey found that only 11% of Americans would be open to using AI for mental health care, just 8% trust it, and nearly half are reluctant.
- A KFF survey revealed that about 77% of Americans are concerned about how their health information would be stored and used by AI systems.
The Risk of Data Breaches Remains Real
Owens highlighted that HIPAA-compliant systems have experienced major data breaches in the past. Quinn expressed a specific fear: that her therapy recordings could be exposed in future breaches.
Even if the system is compliant today, there’s no guarantee it won’t be hacked tomorrow.
The Upside: Massive Time Savings
A study at The Permanente Medical Group reported that AI scribes saved clinicians the equivalent of nearly 16,000 hours of documentation time in one year—a powerful incentive for adoption despite the risks.