On February 25, 2024, an anonymous Russian-language Facebook account published an audio recording of a telephone conversation in which Ukrainian President Volodymyr Zelenskyy is allegedly speaking with an unknown person, asking him to evacuate his parents from the country before the war starts. In the recording, Zelenskyy says that Russia did not invade Ukraine on February 16, but according to partners’ information, it is expected to attack on February 22. Zelenskyy asks his interlocutor to take his parents to Spain or the Caribbean, where the weather is warm. According to the circulated post, the phone conversation took place on February 16, 2022.
The video was originally shared on February 18, 2022, by Ukrainian lawmaker Renat Kuzmin, who cited the Telegram channel “Женщина с косой” [Woman with the braid] as the source.
The telephone conversation, in which Volodymyr Zelenskyy allegedly plans to evacuate his parents, is fabricated and was likely generated using artificial intelligence. Myth Detector, along with its partner organizations, used various tools to establish that this phone conversation never took place.
-
Hiya’s Results
Myth Detector used several tools to determine the authenticity of Volodymyr Zelenskyy’s phone conversation. According to Hiya, an audio detector integrated into InVID & WeVerify, the recording is 73% likely to have been generated by artificial intelligence (AI).
The graph below visually represents the AI generation range per second in the 1:04-minute-long video.
The audio recording was also examined on Deepfakes Analysis Unit’s detection partner, Contrails AI. According to this tool as well, only 37.31% of the recording was likely to be real. Their final verdict concluded that the recording was fake.
-
Hive AI’s Audio Detector Results
Hive AI’s audio detection tool determines whether an audio file is AI-generated. However, due to background noise in the recording, the platform struggled to accurately determine whether the video was AI-generated. The detector divided the file into 10-second segments, producing varied results for each.
Independently from Myth Detector, the organization also analyzed the audio using Hiya to check for signs of a real human voice. According to the Deepfakes Analysis Unit, only 22% of the voice in this specific video was identified as human, making it highly probable that the material was AI-generated.
-
Contrails’ Results
The audio recording was also examined on Deepfakes Analysis Unit’s partner platform, Contrails. According to this tool as well, only 37.31% of the recording was likely to be real. Their final verdict concluded that the recording was fake.
-
WITNESS Team’s Analysis
Myth Detector also reached out to the WITNESS team, which specializes in verifying the authenticity of visual and audio materials. Zelenskyy’s phone conversation was reviewed by two separate teams within WITNESS. The first team did not detect clear AI-generated traces but suggested that the audio might have been processed using a high-pass filter and other manipulation techniques.
To determine the authenticity of the recording, the second team of WITNESS compared it with ten other speeches by President Zelenskyy. They used two voice recognition tools, both of which indicated that the recording was most likely fabricated.
It is worth noting that the audio is of low quality, which makes it more challenging for voice recognition tools to deliver precise results.
The article has been written in the framework of Facebook’s fact-checking program. You can read more about the restrictions that Facebook may impose based on this article via this link. You can find information about appealing or editing our assessment via this link.
Read detailed instructions for editing the article.
Read detailed appeal instructions.
