AI and data use.
Which features use AI, which providers process your data, and how model output is checked before it becomes a result.
Call analysis and generated ad text use Anthropic's models. Call audio is transcribed through AssemblyAI's EU service, and each transcript is deleted from AssemblyAI once Signet has it.
Call analysis is checked against a versioned output contract before it is saved. Output that is malformed, cut short, or cites something not in the transcript is refused and logged, never stored as a result.
Every quote a judgment relies on must appear word for word in the call. Low-confidence results go to a review queue, and a reviewer's correction is saved as a new verdict with the machine's kept beside it.
Below 30 observations a verdict or recommendation reads "needs more data" instead of a conclusion, and a check stops any surface setting its own floor.
AI & Data Use Policy (pending legal drafting)
Not yet settled: Model training
Whether Signet or its model providers may train on customer data is confirmed against each provider's current terms and stated in the AI & Data Use Policy. Nothing in the code controls it.