Experts recently warned that real-time voice deep forgery technology has matured and is bringing new network security risks. With the widespread availability of open source AI tools and affordable hardware, attackers can use artificial intelligence to disguise and imitate anyone's voice in real-time conversations, breaking through previous technical limitations that can only handle pre-recorded content or require long processing times.
New research from cybersecurity firm NCC Group shows that by combining an AI voice model with an ordinary laptop or smartphone, high-quality real-time voice imitation can be achieved with only half a second delay. Operators can initiate voice cloning on a custom web interface with simple operations. This "deepfake voice phishing" attack method can be completed with a lower-configuration graphics card, and even the microphone of an ordinary device can be used to obtain realistic enough effects.
Previous voice deep forgery technology usually requires a long time to train voice data, can only generate pre-recorded clips, and is not suitable for real-time interaction. This breakthrough completely eliminates pauses and unresponsiveness in the voice imitation process, greatly improving attack efficiency and concealment. NCC Group security consultants found in actual tests that when combined with caller ID spoofing, this type of attack can deceive the target almost every time, and the risk of phone voice verification impersonation increases significantly.
Although real-time voice deep forgery is becoming more and more realistic, there are still technical obstacles to the same level of real-time video deep forgery, such as facial expressions and voice being out of sync, which are easy to detect. For example, experts said that one company was even defrauded by AI fake videos and sent laptops to the wrong address, showing that audio and video calls alone can no longer ensure identity verification security.
As artificial intelligence tools become more popular, experts are calling for more sophisticated remote verification methods, such as adding unique structured signals or secret codes to communications, to prevent social engineering attacks caused by deepfakes. Otherwise, both individuals and organizations will face a higher risk of AI counterfeit fraud.