Security boundaries

Qyphera does not collect private keys or seed phrases. Wallet assessments use read-only data. Sentinel applies deterministic rules; AI output cannot override authorization policy.

The development threat model covers tenant isolation, scoped credentials, replay, spending reservations, unavailable providers and input validation. Automated checks do not replace an independent expert assessment.

Production deployment is blocked pending configuration review, threat-model review, provider verification, monitoring, incident response and independent security review. A verified disclosure contact and response process must be established before launch.