A video that appears to show the 35th American president John F. Kennedy criticising the “Israeli leadership” for its “expansionist ideas” has been shared thousands of times online.
But there’s no evidence the clip is real and experts have told us it has likely been made using artificial intelligence (AI).
The clip, which is circulating on Facebook, Instagram and X, appears to show President Kennedy seated at a desk. Some posts claim it is a “never-before-seen clip”.
He supposedly says: “I have made it clear to the Israeli leadership that I do not support their expansionist ideas. The indigenous people of that land have rights that cannot be ignored. We must find a fair solution, not one dictated by any single group.”
Dr Siwei Lyu, an expert in digital media forensics at University at Buffalo, State University of New York, told Full Fact his analysis showed the video was “likely AI-generated”.
He pointed to several visual clues including a dark ‘masking halo’ on the cheek which he said reveals a visible seam where the AI lip-synced movements failed to blend with the rest of the face, as well as an error near the tie, which shifts unnaturally in sync with jaw movement. Dr Lyu also highlighted how the torso is completely static in the clip while the face moves, which he said is “inconsistent with natural human speech”.
We contacted the John F. Kennedy Presidential Library and Museum about this clip and they told us they were not familiar with the video and could not find any audio or transcripts that specifically contained those remarks about Israel.
We also sent the clip to the human rights non-profit WITNESS's Deepfake Rapid Response Force, which had three media forensics teams analyse the visual elements, with one team additionally analysing the audio. All three concluded the video was “likely manipulated”.
One of the teams, Recod.ai, analysed the video using a tool that, given an image of a face, simultaneously learns to decide whether it has been manipulated and to indicate where the manipulation occurred. They divided the video into 92 frames, most of which they said had a high probability of being fake.
Another team, Cauth AI, said there was “a noticeable disconnect between physical cadence and the phrasing of the synthetic audio”, noting that “natural speech involves coordinated body language; [but] here, the physical movements and the audio are completely divorced”. They also said that the lip movements are slightly out of sync with the audio, and the texture of the manipulated mouth area appears unnaturally smooth and blurred.
It’s important to rely on sources that are trustworthy and verifiable before sharing content which you see on social media. Our toolkit and guides on identifying AI-generated content, have tips on how to do this.