Voice fraud detection for banks and call centers.

The voice is now the weakest link

A convincing clone needs only seconds of sample audio, and the scripts are consistent. In executive-impersonation (CEO) fraud, an attacker mimics a leader and instructs a finance team member to move money, as in the 2024 Hong Kong case where a worker was deceived into transferring about 25 million US dollars (reported by CNN). In vishing and family-emergency scams, the same technique targets call-center agents and relatives. The US Federal Trade Commission has warned that scammers now use AI to enhance these schemes, and losses are logged by the FBI's IC3.

The common thread is urgency plus a trusted-sounding voice used to bypass normal controls. Detection is the second line of defense behind process (dual approval, callbacks): once a recording exists, you can prove whether the voice was synthetic.

From flagged recording to filed verdict

The detector reads the acoustic signature of the audio, so it works on cloned and stock voices and names the generator where it recognizes one. It fits a review workflow, not a live switchboard.

Detection, not authentication

We detect whether a voice is machine-generated. That is different from voice biometrics, which verifies a known enrolled speaker as a login factor. The two are complementary: biometrics answers "is this the right person," and we answer "is this a real human voice at all," which is exactly the gap synthetic-voice attacks exploit. Use the detector as an anti-spoofing and post-incident check alongside your existing controls.

We analyze recordings after the fact. We do not tap, monitor, or block live calls, and we do not store your audio beyond 24 hours unless you save a verdict. For the audit trail and citation model, see the API; for how the underlying detection works, see how to detect AI voices. Related reading: spotting AI voice scams and the dangers of AI voices.

Common questions