Every radiology AI product makes one decision before it reads a single scan: where inference runs. Cloud tools send the study off-site to the vendor’s servers. On-device tools keep it in the building. That single choice sets the privacy posture for everything that follows.
The transfer itself is the largest risk. The moment a study leaves your network for inference, the attack surface widens: it transits the public internet, lands on infrastructure you do not control, and is often retained there. Keeping inference on-device removes that transfer entirely, instead of defending it in flight.
The choice also decides data residency and cost. On-device, residency is wherever your hardware sits, which keeps data-protection decisions in your hands; there is no per-study cloud inference dependency and no reliance on connectivity. And it changes nothing about clinical responsibility: with OxRad, the on-device draft is reviewed, edited, and signed by a qualified radiologist on every study.
For the short version, see why on-premise matters for radiology.