Video of Trump appointee calling for Imran Khan's release bears signs of AI

A prominent appointee of US President-elect Donald Trump has publicly called for the release of Pakistan's jailed former leader Imran Khan but footage appearing to show Richard Grenell extolling Khan bore signs of being AI-generated. The circulating clip, viewed hundreds of thousands of times, was an edited version of a speech Grenell delivered during the July 2024 Republican National Convention where he praised Trump and made no mention of Khan.

The video has been viewed more than 150,000 times since it was shared on TikTok on January 3, 2025.

It seemingly shows Grenell -- a former acting director of US national intelligence who Trump appointed in December to serve as a presidential envoy for special missions -- saying, "Ladies and gentlemen, today I want to talk to you about a leader who has become a symbol of justice, integrity and service to the people not only in his country but for the entire world, Imran Khan" (archived link).

"But today, he is imprisoned. Why? Because he speaks the truth. Because he stands for what is right," he appears to say. "We must all come together and raise our voices. Free Imran Khan."

Image

Screenshot of the false TikTok post, captured on January 13, 2025

Pakistan's former prime minister Khan was arrested in May 2023 after being ousted from office and mounting an unprecedented campaign of defiance against the nation's powerful military leaders (archived link).

His detention over graft allegations sparked nationwide unrest, some targeting armed forces installations, and prompting rare prosecutions of civilians in military courts.

Khan's May 2023 detention lasted a matter of days but he was re-arrested three months later and has remained imprisoned since, facing a parade of court cases he claims are politically motivated.

The video was also shared in X posts here and here that racked up more than 140,000 views.

Comments indicated many users believed it was genuine.

"We salute Mr Richard, thank you sir for an honest announcement for the great honest leader Imran Khan," one wrote. Another said, "Thanks Richard for standing with truth and justice."

Grenell had repeatedly called for Khan's release -- posting "Free Imran Khan" on his X account here, here, here and here -- but the circulating video bears signs of being AI-generated (archived here, here, here and here).

RNC speech 

Reverse image searches of the video's keyframes on Google found it originally showed Grenell giving a speech at the RNC on July 17, 2024.

In the speech, published in full by C-SPAN, he praised Trump and did not mention Khan (archived link).

"After 16 years of wars and chaos, the world looked different when President Trump was in the White House," he said. "Russia invaded Ukraine under both Obama and Biden, but Putin did not dare escalate under President Trump."

"Under Trump, China was challenged, the Balkans were more peaceful and the Middle East signed his historic peace deals," he added.

Below is a screenshot comparison of the circulating video (left) and Grenell's original speech (right): 

Image

Screenshot comparison of the circulating video (left) and Grenell's original speech (right)

AI errors

Moreover, elements of the video also indicate it was AI-generated.

At certain points in the clip, Grenell's voice is not in sync with his lip movements. At the-19 second mark, his voice is heard but the video shows a member of the audience whose lips are mouthing the words.

Below is a screenshot comparison of the video in the false posts (left) the original showing the audience member (right):

Image

Image comparison of the falsely shared video (left) with the RNC speech (right)

Siwei Lyu, director of Media Forensic Lab at the University of Buffalo, analysed the video and pointed out other errors (archived link).

He told AFP on January 13: "Noticeable artefacts are present around the mouth and eye areas. The eye colour does not match and the overall expression of the eyes appears unnatural."

While there is no fool-proof method of spotting AI-generated media, identifying visual inconsistencies can help, as errors still occur despite the meteoric progress in generative AI.