Insurance answers,
on the phone, in seconds.
Aria is a voice AI that patients can call to check what their plan covers, whether a procedure needs prior authorization, and where their claim stands — with identity verification and PHI guardrails built in.
How a call works
Three deterministic steps before any personal health information leaves our systems.
Patient calls
Twilio routes the call to Aria. We verify the request signature on every webhook — nothing else can reach the agent.
Identity is verified
Aria collects the caller's name and date of birth. No plan details, no claims, no diagnoses are shared until the match succeeds.
Coverage answered
Aria uses narrowly-scoped tools — insurance, prior auth, claims — restricted to the verified patient. Every turn is transcribed.
Guardrails, not vibes.
Voice agents leak PHI when identity verification is a prompt, not a gate. Aria enforces it server-side, then a second model reviews every reply.
- Server-side identity gatePHI tools reject any request whose patient ID doesn't match the verified session — even if the model tries.
- No medical adviceAria refuses treatment or diagnostic questions and refers callers to their clinician.
- Injection-hardenedCaller speech is wrapped as untrusted data. A classifier flags jailbreak attempts, out-of-scope drift, and possible PII leaks — visible in the admin dashboard.