A recording of US Vice President JD Vance commenting that Musk is “transforming into a great American leader” shows that AI sounds are becoming increasingly difficult to detect.
Last weekend, an audio file went viral on social media, attracting millions of interested users. In it, Vice President JD Vance “said” Elon Musk was “posing as a great leader” and making the government worse.
US Vice President JD Vance. Photo: The White House
According to 404Media , on TikTok alone, the recording was played over two million times with 8,000 comments but was not labeled by AI. Similarly, on platforms like YouTube and X, the number of views reached hundreds of thousands.
Late on March 23, William Martin, Mr. Vance’s communications director, spoke out on social network X: “The recording is 100% fake, definitely not the Vice President’s words.”
There are some AI-powered audio detection tools out there, but AI is getting better and better, making it harder to identify. Reality Defender, a company that makes software to detect AI-generated misinformation, also rates audio files as “likely fake” rather than saying they are fake.
“We fed the audio clip through multiple deep learning models and concluded it was likely fake,” a Reality Defender spokesperson said. “Noise and echoes were added in a sophisticated way to mask the quality of the audio produced by the deepfake, making it difficult to detect.”
TikTok, YouTube and X have not yet commented.
According to 404Media , the audio is “completely believable” because the voice sounds exactly like Vance. The static is very similar to secret recordings that have been made of other politicians in the past. The background noise also makes it difficult for deepfake detectors to recognize the fake audio.
The fact that even advanced tools have difficulty determining the authenticity of the audio file above has caused concern among experts in the context of deepfake. Deepfake is a combination of deep learning and fake, using AI to analyze a person’s gestures, facial expressions, and voice, thereby recreating and editing to create realistic photos, videos, and sounds.
This type of AI is now easily accessible to ordinary users. With just a voice file and a text input, free online tools can create recordings with arbitrary content. Consumer Reports looked at six popular AI voice transcription products and found that there are no effective measures to prevent users from misusing AI.
Last week, the European police agency Europol warned that AI is contributing to the acceleration of crime. “AI can automate criminal activity, making it more scalable and harder to detect,” the organization said.
Evan Dornbush, a former cybersecurity expert at the US National Security Agency (NSA), told Forbes that AI can help scammers create credible messages faster, and at a lower cost.
According to ZDnet , users can still detect deepfake audio through intonation. If the voice is “monotone, slurred, and unnatural,” it could be AI audio. Additionally, strange noises, or “too clear like a recording studio” can also be suspect. However, with technological advances, AI is gradually overcoming this drawback every day.
The Better Business Bureau offers this rule: “If something sounds too good to be true, be skeptical.”
