The following vulnerabilities are found in voice biometrics:
Replay: The biggest concern is the replay attack. Hackers might attempt to gain unauthorized access to a voice authenticated system by playing back a pre-recorded voice sample from an authorized user. Need to implement proper anti-replay/ spoofing measures.
Voiceprint re-enrollment (Social Engineering): The malicious user claims to the contact center agent that they are unable to authenticate with their voice, and that their voiceprint needs to be re-enrolled. If the agent complies, a fraudster can be enrolled in the system and be provided with access to a legitimate account.
Brute Force attack: This attack consists of a fraudster calling the IVR or call center numerous times until their voice is mistakenly accepted by the voice biometric system as belonging to a legitimate account holder. Vulnerability testing conducted on deployed voice biometric systems indicates that the rate of a success of a brute force attack is between 0.1% and 0.5%.
Mitigation:
Replay attacks-
Any voice identification solution needs to include measures to detect replay attacks.
-Voice biometrics should be able to tell the differences between real and fake users
-Anti-spoofing is the key.
-Challenge Response Mechanism
Text-Prompted Authentication: In text-prompted mode users enroll by repeating a set keywords (digits, places, names, etc). Verification requires the user to repeat a randomly generated sequence of a subset of those keywords. This mitigates the above threat, as the fraudster will not have a recording of the legitimate account holder’s voice speaking the random passphrase.
Text-Dependent Authentication with A Passphrase: Rather than having a universal phrase that an attacker can easily gain knowledge of, users can enroll with their own secret phrase. Users are then responsible for keeping their phrase secret (and remembering it). The system does not prompt the user to speak the specific phrase. Instead it asks them simply to repeat their secret phrase, making it difficult for an attacker to know what set of words to record to execute a replay attack.
Voice re-enrollment-
An agent can verify that the caller has recently enrolled and has not been able to verify. A caller that requests a voice biometric re-enrollment that has successfully authenticated previously is most likely either a fraudster, or does not need to be re-enrolled.
Brute force-
Very similar to classic brute force attack: Block the caller after pre-determined unsuccessful login attempts. If there are three concurrent failed authentication attempts on a single account, that account can be locked to minimize the probability of a successful attack.
Replay: The biggest concern is the replay attack. Hackers might attempt to gain unauthorized access to a voice authenticated system by playing back a pre-recorded voice sample from an authorized user. Need to implement proper anti-replay/ spoofing measures.
Voiceprint re-enrollment (Social Engineering): The malicious user claims to the contact center agent that they are unable to authenticate with their voice, and that their voiceprint needs to be re-enrolled. If the agent complies, a fraudster can be enrolled in the system and be provided with access to a legitimate account.
Brute Force attack: This attack consists of a fraudster calling the IVR or call center numerous times until their voice is mistakenly accepted by the voice biometric system as belonging to a legitimate account holder. Vulnerability testing conducted on deployed voice biometric systems indicates that the rate of a success of a brute force attack is between 0.1% and 0.5%.
Mitigation:
Replay attacks-
Any voice identification solution needs to include measures to detect replay attacks.
-Voice biometrics should be able to tell the differences between real and fake users
-Anti-spoofing is the key.
-Challenge Response Mechanism
Text-Prompted Authentication: In text-prompted mode users enroll by repeating a set keywords (digits, places, names, etc). Verification requires the user to repeat a randomly generated sequence of a subset of those keywords. This mitigates the above threat, as the fraudster will not have a recording of the legitimate account holder’s voice speaking the random passphrase.
Text-Dependent Authentication with A Passphrase: Rather than having a universal phrase that an attacker can easily gain knowledge of, users can enroll with their own secret phrase. Users are then responsible for keeping their phrase secret (and remembering it). The system does not prompt the user to speak the specific phrase. Instead it asks them simply to repeat their secret phrase, making it difficult for an attacker to know what set of words to record to execute a replay attack.
Voice re-enrollment-
An agent can verify that the caller has recently enrolled and has not been able to verify. A caller that requests a voice biometric re-enrollment that has successfully authenticated previously is most likely either a fraudster, or does not need to be re-enrolled.
Brute force-
Very similar to classic brute force attack: Block the caller after pre-determined unsuccessful login attempts. If there are three concurrent failed authentication attempts on a single account, that account can be locked to minimize the probability of a successful attack.
Comments