We do not train AI on your data

Session transcripts, typed messages, voice audio, stored memories, emotional snapshots, journal and check-in content, and anything else you tell Ophie. None of it is used to train, fine-tune, or otherwise adjust the weights of any model — ours or a provider’s.

Most providers that process your data on our behalf are contractually restricted from training on it. A few run model-improvement programs that use inputs unless we opt out, and a few say nothing either way in their published terms — we name both individually rather than claim a blanket restriction. Each provider, what it receives, and its training posture is listed on the subprocessors page.

We evaluate and improve Ophie using synthetic conversations — scenarios written by our team and played out between AI models, with no real user involved. Clinical reviewers read those synthetic transcripts, not yours.

Where we have experimented with fine-tuning, the training data was this synthetic material. If a fine-tuned model ever reaches production, its data sources will be summarized on this page first.

Before any training on user data could begin, all of the following would happen first:

• We would ask for your separate, specific opt-in. It would not be bundled into the Terms, the Privacy Policy, or any other consent, and declining would not reduce your access to the service.
• We would publish a plain-language training-data summary on this page before training starts, covering the AB 2013 items below.
• We would give advance notice to affected users where the law requires it, and update the Privacy Policy and Consumer Health Data Privacy Notice in the same release.

California’s AB 2013 sets out what a training-data disclosure has to contain. If we ever owe one, it will be published here and will address each item:

• The sources or owners of the data sets
• A description of how the data sets further the intended purpose of the system
• The number of data points in the data sets
• Whether the data sets include any data protected by copyright, trademark, or patent
• Whether the data sets were purchased or licensed
• Whether the data sets include personal information or aggregate consumer information
• Whether the data sets were cleaned, processed, or otherwise modified, and for what purpose
• The time period during which the data were collected, and whether collection is ongoing
• Whether the data sets include synthetic data generation