Google Built an AI System to Detect AI Impersonation Calls Before They Reach You

Google built a system to detect ai-generated voice calls. The system was trained on synthetic voice samples. Those samples were generated by another Google system. The training data came from the same company that built the detection model. The system is therefore trained on what Google can generate.

This is the standard arrangement where the problem and solution originate from the same source. Google's detection system cannot detect what it was not trained on. Other companies' voice synthesis will sound unfamiliar. The unfamiliar may trigger false positives or false negatives. Control of the training data means control of the detection blindness.

The system will be deployed. It will stop some calls. It will miss others. Google will publish statistics showing it works. The statistics will be accurate. Everything is under control because everything is under google.