AI voice cloning: how a phone verifies whether a real contact is calling

The phone rings, “Mom” lights up on the screen, and a familiar voice comes through the speaker. He sounds frightened, talks about an accident, and urgently asks for money. This is exactly what voice-cloning scams look like, and the problem is that people may have a harder time recognizing a voice than they think. What can a phone do today—and what can’t it do?

Why Is a Cloned Voice So Effective?

Artificial intelligence can use a speech sample to create a synthetic voice that mimics the timbre, pace, and intonation of a specific person. All the scammer then needs to do is conduct the call under pressure and not give the victim time to think.

Research published on the arXiv server showed that people find it difficult to reliably detect a synthetic voice. In the experiment, the cloned voice was perceived as identical to the original voice in approximately 80% of cases, and correctly identified as artificial in only about 60% of cases. Relying on one’s own hearing is therefore a weak defense.

What New Phone Features Actually Do

In light of these findings, it’s tempting to believe that a phone “listens” to a call and can tell whether a human or a machine is speaking. The actual solutions emerging today work differently, and it’s important not to confuse them.

Device verification, not voice analysis

On June 2, 2026, Google announced a feature for Android that detects fake calls. It’s a kind of “digital handshake” between two devices. When a contact calls you and both of you are using the Google Phone app, their device sends a silent confirmation signal to yours, verifying that the call is actually coming from their phone. If the signal is missing, your phone will attempt to verify this directly on the original contact’s device. If that device responds that it is not currently making a call, a warning will appear on the screen advising you to end the call.

This feature does not analyze the sound of the call at all. It does not look for breathing patterns or acoustic anomalies in the voice, as is sometimes claimed. It only verifies whether the call is actually coming from the device that claims to be the contact.

Analysis of Call Patterns

There is a separate feature that monitors typical fraudulent patterns—such as urgent requests for money—locally on the device during calls with unknown numbers. This is a different mechanism than device verification, and according to available information, it is designed for calls from unknown numbers, not for recognizing the synthetic voice itself.

Where Are the Limitations

The current conditions are relatively narrow, so they cannot be relied upon as the sole form of protection.

• The feature is gradually rolling out to devices running Android 12 and newer, starting with Pixel phones, and requires the Phone, Contacts, and Messages apps from Google.
• Both phones—yours and your contact’s—must use Google’s Phone app.
• By the very nature of the feature, it won’t help if a scammer doesn’t impersonate a saved contact but calls from a new number claiming to have “lost their phone.” Device verification won’t detect such a scenario.

For iPhone users, we were unable to find a verified equivalent of this specific feature while preparing this article. Therefore, we do not present information about upcoming security features in iOS as a confirmed fact.

How to Protect Yourself Regardless of the Technology

Family Passcode

Agree on a secret word or question with your loved ones that only your family knows. If someone calls you in a panic and asks for money, ask for the password. A synthetic voice won’t know it unless you’ve shared it somewhere.

Call Back

End the call and call the person back using the number you have saved. If at all possible, verify the situation through another family member or a different channel.

Don’t give in to pressure

A typical sign of a scam is time pressure: an accident, an arrest, a debt collection notice, or a demand that you not tell anyone. If someone is pressuring you to transfer money immediately, that’s a reason to slow down, not speed up.

Conclusion

Technology for protecting against scam calls is shifting toward verifying devices rather than detecting “glitches” in the voice, which makes sense since synthetic voices are becoming increasingly convincing. For now, however, this protection has limited coverage. Therefore, the most reliable defenses remain a prearranged password, a callback, and a cool head.