AI Assistant Eva Flags Deepfake Voices in Real Time
Megafon’s virtual secretary can detect synthetic speech during calls and warn users about potential fraud.
In Russia, Megafon’s virtual assistant Eva can alert users during a phone call if the voice on the other end is synthetic or generated by AI. The system analyzes audio across multiple parameters, while a multi-factor model identifies signs of synthesized speech. Eva can distinguish between benign though annoying automated calls, such as delivery notifications, and malicious deepfake attempts, which often mimic voices of relatives, managers or public figures to deceive users.
Behind the Call
If the assistant detects signs of synthetic speech, it immediately notifies the user. At the same time, the system checks for indicators of fraud. If such signs are found, Eva automatically terminates the call. The AI operates in real time: when risky patterns emerge, the assistant warns the user and disconnects the call.
Not all synthesized speech is malicious. Automated calls from delivery services or notification systems are not considered a threat. Eva distinguishes these from harmful deepfakes and does not interrupt such interactions.
