The UK is facing a “democratic emergency”, a leading think tank has warned weeks ahead of a crucial by-election, after new polling revealed around 16.5 million UK adults saw political deepfakes in the month before the local elections.
Almost one in three (30 per cent) voters said they had seen a deepfake or AI-generated video, audio clip or image about an election candidate or politician online in the lead-up to last month’s elections, the shocking new polling showed.
The polling of 2,005 adults was conducted by Opinium for the cross-party think tank Demos between 30 April and 6 May this year – immediately before local and devolved elections across the UK.
The warning comes just two weeks before voters go to the polls in Makerfield for a contest that could decide Britain’s next prime minister if Andy Burnham wins the seat and decides to challenge Sir Keir Starmer, which he is widely expected to do.
Deepfakes are digitally created and altered content, often in the form of fake images, videos and audio recordings.
Around one in six people (16 per cent) said they had encountered political deepfakes more than five times during that period, suggesting that a significant minority of users are being exposed to this content at very high levels.
The polling showed that, when it comes to UK politicians, Labour and Reform leaders were most often reported to have been “deepfaked” – which is significant for the upcoming Makerfield by-election as the race is expected to be a close fight between those two parties.
The findings come as the Electoral Commission launches a new deepfake detection pilot intended to improve identification and map the scale of the problem. However, findings from the pilot are not expected for at least six months.
Demos has called for the government to “move faster to establish clear rules and accountability” for deepfakes, urging ministers to “use the Representation of the People Bill – which is already underway – to introduce meaningful protections for the public before the next general election”.
The think tank previously proposed amendments to the Representation of the People Bill to address AI-generated election misinformation, including clearer legal responsibilities for platforms and developers – proposals which were not taken up by the government.
Chi Onwurah, chair of the Science, Innovation and Technology Committee, agreed that “stronger safeguards are needed to protect voters from online misinformation”.
She told The Independent: “Reports of growing numbers of political deepfakes ahead of elections is deeply alarming, and it’s clear that stronger safeguards are needed to protect voters from online misinformation.
“My committee has repeatedly raised the risks posed by AI-generated deepfakes with the big tech companies, and we weren’t satisfied with their response. It’s clear that protections are not working as intended.
“Deepfakes can do untold damage to individual lives and to the integrity of our democratic systems, eroding confidence In the electoral process itself. At a time where hostile actors like the Kremlin are actively working to undermine our democracy, complacency isn’t an option.”
Demos’s research also found that 39 per cent of respondents were unsure whether they had seen a deepfake at all, which the think tank argued showed a “concerning lack of public confidence in discerning the truth in visual content they see online related to elections”.
More of the public said they were not confident they could identify a deepfake (43 per cent) than said they were confident (38 per cent), the survey also showed.
The most commonly identified subjects of political deepfakes were Donald Trump – with 45 per cent of people saying they had seen a deepfake of him – Sir Keir (36 per cent) and Nigel Farage (27 per cent). Meanwhile, a smaller minority reported seeing deepfakes of Zack Polanski (10 per cent) and Kemi Badenoch (8 per cent).
Polling suggested much of the content was overtly damaging to the profiles of those represented. Among respondents who had seen political deepfakes, six in 10 (56 per cent) said the content portrayed the subject negatively, including 28 per cent who described the content as “very negative”.
The polling also found significant public concern about the impact of AI misinformation on democracy, as 42 per cent said they were worried about fake videos or deepfakes of candidates and MPs having an impact on the 7 May local and devolved elections, while just 23 per cent said they were not worried.
Azzurra Moores, associate director of information ecosystems at Demos, told The Independent: “Political deepfakes are no longer a future threat, they are already flooding people’s social media feeds.
“Our polling shows millions of people say they are now encountering AI-generated political content online, often repeatedly and usually in a negative context. At the same time, many voters are unsure how to discern the truth from the content they are seeing.
“That combination of widespread exposure and low public confidence in spotting deepfakes creates serious risks for trust in democratic debate, setting the stage for a democratic emergency in the UK.
“As it stands, generative AI is evolving faster than our democratic protections. The UK cannot afford to drift into the next general election without stronger safeguards in place.
“The evidence of harm is mounting and the opportunity to act is narrowing. The government has a clear chance now to protect both the public and our democracy before this threat becomes a full-blown crisis. It must not be wasted.”
A government spokesperson said: “Deepfakes can sow division and manipulate public opinion, posing a growing threat to public trust – and we are taking action.
“Under the Online Safety Act, social media platforms must take measures to tackle illegal content – including fraud by false representation – or face enforcement action.
“We know the threat is evolving rapidly, which is why during election periods we stand up specialist teams to monitor risks and respond in real time to online misinformation.”
