AI Training Transparency

No.Daylogue does not train, fine-tune, or otherwise use any user check-in entries, voice transcripts, journal vault content, mood metrics, or AI-generated narratives to train, fine-tune, evaluate, or improve any AI model — whether ours or a third party's.

• Anthropic Claude family — accessed exclusively through AWS Bedrock under a zero-data-retention configuration. Anthropic does not have access to inputs or outputs.
• Deepgram — voice-to-text only, under a zero-retention contract. No training on customer audio or transcripts.

All inference runs server-side under our AWS account. No cross-model data sharing. No federation to consumer accounts.

Daylogue has not developed, fine-tuned, or modified an AI foundation model. We use first-party prompts and orchestration over third-party models. Because Daylogue does not train its own model, the AB 2013 dataset-summary requirement applies to us only as a downstream user of Anthropic and Deepgram, both of whom publish their own training-data summaries:

• Anthropic training data summary: anthropic.com/legal/training-data
• Deepgram model training disclosure: deepgram.com/legal

In compliance with EU AI Act Article 50 and California SB 1001, Daylogue discloses AI involvement at every check-in session. The disclosure appears before the first text exchange, voice session, or SMS conversation. Disclosure language is recorded in the product copy register and reviewed quarterly.

Operational logs are scrubbed of personal content before storage and are never used to train models. Sentry error reports use a configured PHI scrubber. Analytics is event- and aggregate-level only.

This statement is reviewed at least annually and on any material change in AI vendor, training posture, or model configuration. Material changes are announced on this page and on the Trust & Security page.

privacy@daylogue.io

Daylogue is not therapy and is not a replacement for professional mental health care.