The integration of artificial intelligence into everyday life has reached a point where patients can now record therapy sessions using consumer-grade wearable technology, raising significant ethical and privacy concerns for mental health professionals. Victoria Song, a senior reviewer at The Verge, spent three weeks testing the Amazon Bee Frontier, a device designed to capture ambient conversation, transcribe it, and summarize it using AI. Her experiment, which involved bringing the device to her long-term therapy sessions, highlights a growing trend where third-party algorithms interpret sensitive personal data without the clinician's knowledge or consent. [1]
Song’s experiment began with full transparency and permission from her therapist, with whom she had worked for 11 years. However, the experience quickly revealed the limitations and potential harms of such technology. On the second day of the study, Song forgot to deactivate the device during a session that she described as cathartic, involving laughter and jokes. Following this session, the AI generated a 'fact' based on a real but deeply personal childhood trauma related to domestic violence. The device labeled the conversation as 'intense, reflective, emotionally raw' and described Song as 'deeply unsettled.' [1]
The discrepancy between Song’s subjective feeling of lightness after the session and the AI’s negative characterization sparked a profound doubt about whose interpretation of the therapeutic encounter was correct. The broader assessment of her nearly month-long trial was unsettling: the summaries were accurate only about 75% of the time, while 15% were taken out of context, and 10% were directly erroneous. Despite warnings that the AI could make mistakes, these outputs were presented as factual records. The device also inadvertently recorded her annual medical review because it remained active under her clothing. [1]
For clinicians, the most alarming aspect is not the summary itself, but the existence of a third, AI-generated narrative that lacks professional secrecy obligations. The therapist pointed out three major risks: the erosion of patient trust, the potential for misinterpretation of sensitive data, and the commercialization of private moments. The irony was compounded when the AI suggested Song buy nail glue from Amazon as a task, turning a therapeutic space into a commercial one. [1]
The debate around recording therapy sessions has traditionally focused on whether the psychologist should record; now, the power dynamics shift when the patient brings an AI listener into the room. Ethical guidelines offer some protection, but they were not designed for this scenario. The Spanish Code of Deontology for Psychologists emphasizes respect for client intimacy (Article 39), professional secrecy (Article 40), and the need for prior consent for the presence of third parties (Article 47). While Article 47 was intended for trainees, the analogy holds: an algorithm processing data on external servers is a third party that does not adhere to professional secrets. [1]
From a legal perspective, data protection laws like the GDPR (EU), the Federal Law on Personal Data Protection in Possession of Individuals (Mexico, 2025), and Law 25.326 (Argentina) classify health data as sensitive. However, these laws often exclude processing for 'exclusively personal or domestic purposes.' When a patient records a session for themselves, they may fall outside the strict protections of data protection regulations, leaving the resolution of conflicts to the therapeutic agreement rather than legal mandates. [1]
Professionals are advised to address this issue proactively. Rather than reacting with alarm, psychologists should include a specific clause in their informed consent forms regarding AI wearables and recording devices. This clause should stipulate that no audio or video recording is permitted without prior mutual agreement. If a patient wishes to record, the therapist should discuss the purpose, the device used, where the data is stored, and who processes it. Alternatives, such as providing a written summary of guidelines after the session, can be offered. [1]
- Categories:
- Technology
- Entities:
