Zero-data-retention
What zero data retention means when AirNote uses AI—and why sensitive therapy content does not become model memory or a provider archive.
2 min read
Therapy depends on confidence that intimate information has a clear purpose and a clear boundary. “Zero data retention” is an AI infrastructure term, but it answers a simple clinical question: once an AI provider has completed a task, does it keep the material it processed?
For AirNote’s AI generation workflow, the answer is no. The relevant text is processed for the requested task, the draft is returned, and OpenAI does not retain the prompt or model response.
Inference is processing, not memory
When a therapist asks AirNote to generate a Note, document or answer, a model receives the information needed for that request and calculates a response. This step is called inference.
Inference does not add the client’s story to the model’s memory or rewrite the model around that session. OpenAI’s API data is not used to train its models by default, unless a customer explicitly opts in. AirNote does not opt in.
What zero data retention changes
With ordinary API settings, OpenAI may create abuse-monitoring logs that contain prompts or responses and keep them for up to 30 days. Zero Data Retention, or ZDR, is a stronger, approved control. OpenAI describes ZDR as a promise not to retain prompts or model responses after a request is processed.
AirNote uses a covered, ZDR-enabled OpenAI API project and a compatible text endpoint with storage disabled. Only the relevant text—not raw session audio—is sent. Once the response has been returned, there is no provider-side copy of that clinical content left to retrieve later.
Why this matters in therapy
A therapy transcript may refer to trauma, relationships, health, risk and other profoundly private parts of a person’s life. Keeping fewer copies of that material is data minimisation in its most practical form: reduce the number of places sensitive content can remain.
When a therapist intentionally requests a text-based AI task, zero data retention places a clear boundary around that temporary processing step.
The simplest way to think about it is that the model is a momentary processor, not a filing cabinet. It receives what it needs to complete one run, returns an editable draft for clinical review, and keeps no clinical content to remember later.
