Audio of President Trump criticising Keir Starmer is fake

A video circulating on social media appears to feature an audio recording of US President Donald Trump criticising the UK Prime Minister Sir Keir Starmer over aid to Ukraine, energy costs and jobs. 

But the audio has been generated by AI and overlaid onto a video interview Mr Trump did with podcaster Theo Von on 21 August 2024. 

Neither Mr Trump or Mr Von discuss the UK or Ukraine in this video, but instead focus on US healthcare reform, border security and President Trump’s reasons for avoiding drinking and smoking.

How did we know this was AI-generated?

We contacted Professor Hany Farid, who specialises in digital forensics at the University of California, Berkeley, and is Chief Science Officer at GetReal Labs, a cybersecurity company focused on preventing malicious threats from generative AI. 

He told us this is “an obvious—and not particularly well-executed—fake”. 

“The audio is AI-generated and at the beginning of the video there is no effort to sync Trump’s video to the audio. In addition, the voice has tell-tale signs of AI-generation in terms of cadence and intonation, a model we have trained to distinguish real from AI-generated voice flags it as AI-generated.”

An expert from audio production company Music Radio Creative also confirmed this is AI, noting “subtle hints on how the audio flows” and how the mouth moves. 

The UK’s deal with Ukraine

While we haven’t fact checked all the claims made by the audio deepfake, it makes at least one claim that is false: that the UK government “just signed a deal to send 300 billion over the next 100 years”.

The UK has agreed to provide Ukraine with annual military assistance of no less than £3 billion a year until 2030/31 “and for as long as needed to support Ukraine” according to the government. 

Misinformation spreads quickly online so it’s especially important to consider whether something is likely to be genuine before sharing.

We’ve written a guide on how to spot deepfake videos and AI audio which can help you identify content that has been edited or created to mislead online.