Claim: A viral video shows US President Donald Trump being evacuated by his security detail from the White House Correspondents’ Dinner.
Fact: The video is likely AI-generated.
On 28 April 2026, Facebook user Naimat Khuhro posted a video with the following caption in Sindhi:
ايران تي پَنجاھ پَنجاھ مڻن جا ھزارين بم ھڻندڙ آمريڪي صدر ٽرمپوءَ جو ھِڪ ھوائي فائر تي ساھ پيو نڪري، ۽ ڊرامو اِيئن پيو ڪري، ڄڻ ته پيٽ ۾ گولي لڳي اَٿس/ حملو ناڪام ٿيڻ تي شھباز شريف کي خُوشي ٿِي، ۽ اسانکي ڏاڍي تڪليف پُھتي آھي/ ڪاش! حملو ڪندڙ ماستر پَنھنجي مقصد ۾ ڪامياب ٿئي ھا تہ جيڪر ڪروڙين اربين انسانن جي ھِن وحشي قاتل ۽ دَجال مان جان آجي ٿِي پَوي ھا!
“[Translation: US President Trump, who dropped 50,000 bombs on Iran, is dying of a single air strike, and the drama is going on as if he was shot in the stomach/ Shahbaz Sharif is happy that the attack failed, and we are in great pain/ I wish! The master attacker had succeeded in his goal, if only the lives of millions of people had been saved by this brutal murderer and antichrist!”
The video shows US President Donald Trump looking shocked as he is escorted out by his security detail after a second assassination attempt.
White House Correspondents’ Dinner
On 26 April 2026, a shooting incident occurred during the White House Correspondents’ Association dinner at the Washington Hilton hotel. The suspect, later identified as Cole Tomas Allen, was armed with a shotgun, handgun, and knives. Secret Service Spokesman Anthony Guglielmi said that Allen opened fire near the main screening area at the entrance of the event. He was then tackled and arrested.
US Acting Attorney General Todd Blanche told NBC News that the suspect was likely targeting US President Donald Trump and officials in his administration.
Footage by CNN shows the attendees taking cover behind their tables as the security personnel rushed to cover Trump and first lady, Melania Trump, who bent down behind the dais before being escorted out by the officers. Security agents also provided cover for other cabinet officials including Vice President JD Vance, Secretary of State Marco Rubio, Health Secretary Robert F. Kennedy Jr., and Interior Secretary Doug Burgum.
After an hour of the incident, Trump announced on his Truth Social account that the “shooter has been apprehended”.
Allen is a 31 year-old teacher and engineer from Torrance, California. NBC News claims that a senior official from the US administration has provided them with a transcript of Allen’s writings which show that the suspect had sent a message to his parents, colleagues, and students, before the attack. He had apologised for what he was about to do as he believed that it was his “duty” to target Trump administration officials.
Fact or Fiction?
Soch Fact Check reverse-imaged the thumbnail of the videos and photos to see if it matches the official videos that were released by news reports on the date of the incident.
When analysed alongside with the claim, we found there are no close-up videos of security escorting President Donald Trump away from the stage. Two official clips were released, one is a video of the US President and First Lady Melania Trump engaged in conversation with mentalist Oz Pearlman, as shots are heard being fired.
The next clip shows Donald Trump being escorted out with security gathered around him, as US Police stomped the stage.
The only close-up shot taken on 25 April 2026 shows Vice President JD Vance, being taken away from the stage by his security detail. These are three main clips circulating social media.
Soch Fact Check verified the video by running it through Deepware AI Scanner, which assesses videos with different AI-based scanners that check for synthetic manipulation and produce a probability score.
Running it through the Avartify scanner yielded the highest probability, indicating the video is 90% likely a deepfake. Avartify is an AI-powered deepfake detection tool, specifically focused on using facial animation to analyse videos for altered facial animation within a video, or even called as a puppetry tool. This can be seen in the beginning of the video, where Trump’s expression seems to be motionless and stationery with the shocked expression not moving at all..
Furthermore, the Seferbekov scanner has been designed to detect inconsistencies between the person’s head movement and the surrounding background pixels. This scanner saw 56% of suspicious activity, which also suggested AI alteration.
However, results from the other two scanners Deepware at 29% and Ensemble at 43% of a deepfake not being detected. Therefore, results from Deepware Scanner were inconclusive.
The results are pictured below:
Results from Deepware Scanner
Soch Fact Check then ran the video through DeepFake-O-Meter for further verification. DeepFake-O-Meter gave us the report below. Even though the summarized result is Mixed Signals, all scanners giving a high to moderate score can be seen as an indicator that the video was in fact a deepfake.
The report shows five scanners that show a high score of deepfake, this included AVSRDD, DSP-FWA, Altfreezing, XCLIP, and LSDA. All detected different things in AI-generated videos for example. The percentage and their functions have been stated below for clarity:
- AVSRDD was at 100% that is an audiovisual deepfake detection that takes into account the correlation of visual and audio cues by comparing discrepancies between audio and video.
- DSP-FWA(Deep Spatial Pyramid – Face Warping Artifacts) scored 95.5% identifies visual distortions or “warping” artefacts that are typically left as a residue in the swapped or manipulated faces in deepfake videos.
- AltFreezing gave us a probability of 93.4% and is a technique identifying abnormal facial motion or frame freezing patterns that exist in AI-generated videos.
- XCLIP is a transformer-based video understanding model for deepfake detection that captures relationships across frames to detect synthetic patterns and temporal inconsistencies that scored 90.4%
- LSDA is known for a deepfake detection based on spatial anomalies in manipulated media that identifies subtle inconsistencies between frames which detected manipulation of 67.6% in the video.
Result from DeepFake-O-Meter
Also seen below are four scanners that gave moderate scores, which add to the suspicious nature of the video. These four scanners are stated below:
- TALL is a tool made to detect temporal inconsistencies in long videos by analysing action and motion patterns across frames, and it detected 59.1% of inconsistencies across the video
- SBI is an AI-detection model that trains models using self-blended images to help detect subtle face manipulation artefact scoring an 53.9%
- FTCN is a detection framework identifies irregularities in facial movements over time with a probability of the video of 48.5% here we can see how
- Wav2Lip-STA is a detection tool that checks the lip-sync accuracy by comparing spoken audio with mouth movements in vide0 and it detected a probability of 42.4%
Results from DeepFake-O-Meter
When reviewing the forensic analysis done by AI-detection tools, and official coverage of the incident from credible international news outlets from the day of the Correspondent’s dinner, it can be established that the video is not authentic, and likely AI-generated.
Virality
Soch Fact Check found the claim circulating here, here, here,here, here, here, here, here, and here on Facebook.
On Instagram, it was shared here and here.
On Youtube, it was found here.
On Tiktok, it was found here.
Conclusion: The claim is false as the video is likely AI-generated. __________________________________________________________________________________________
To appeal against our fact-check, please send an email to appeals@sochfactcheck.com