Election officials are working with the Home Office on a fast-tracked pilot to counter deepfakes targeting candidates in Scottish and Welsh elections.
Before election campaigns start in late March, the Home Office and the Electoral Commission in Scotland anticipate that software that can identify deepfake videos and images created by artificial intelligence will be ready.
Although she conceded that it could not always guarantee 100% assurance, Sarah Mackie, the commission's chief in Scotland, stated that authorities would contact the police, the candidate in question, and the public if the software identified a hoax video or image.
According to her, they would then demand that the social media site remove the information. However, the commission also wants legally enforced "takedown" powers that would compel media companies to remove false content because such action is now voluntary.
“What we don’t have at the moment, and what we want, is called takedown powers – where we approach social media companies and demand something is taken down,”
Mackie said.
Although there are no documented instances of deepfakes appearing during UK election campaigns, their use has increased in elections overseas, a trend that has accelerated significantly with the proliferation of free AI image-generation tools.
State-sponsored fake social media accounts from nations like North Korea, Iran, and Russia have frequently targeted British elections and referendums with the intention of escalating controversy or spreading discontent.
Speaking at a pre-election briefing for journalists in Edinburgh, Mackie stated that the commission was also collaborating with the Scottish parliament and police on a "safety and confidence" project to assist women and candidates from Black, Asian, and minority ethnic backgrounds who face harassment or abuse based on their race or gender.
According to a 2022 research, about half of the female candidates had been abused, and many of them indicated they would never run for office again. Additionally, candidates from minority ethnic backgrounds said that the abuse undermined diversity at Holyrood by making them too afraid to run for office again.
According to Mackie, if AI-driven and pornographic "undressing" technology, especially that produced by Elon Musk's Grok AI platform were utilized during an election, it would fall under that category and be reported to the authorities.
Senior Westminster MPs have called for immediate action from the government and media regulator Ofcom in response to criticism of Musk's X and Grok platforms for failing to remove toxic, pornographic, and fraudulent content.
According to Mackie, neither the Home Office nor the Electoral Commission had a clear legal duty to control deepfakes during elections, but they both needed to see what they could do.
If the trial initiative was successful, she suggested, it might be implemented for all UK elections.
She said:
“We don’t regulate campaigning but there is an empty space here where it’s a little bit like there’s lots of regulations surrounding the edge of the ring.
So what we are doing is just jumping into the centre of the ring by sort of saying, well, let’s see what we learn from this and then share it with the other people.”
According to a Home Office representative, social media businesses are mandated by the UK's Online Safety Act to eliminate illegal content and stop the dissemination of misleading information that could endanger people's mental or physical health. It said:
“Protecting elections from the threat of sophisticated deepfakes is vital to maintaining public trust in our democratic system.
This pilot will help to detect and tackle deepfake material and protect the public from the impact of disinformation.”
How will the pilot detect AI generated content on social platforms?
The airman design detects AI- generated content on social platforms through real- time monitoring of TikTok, X, Facebook, Instagram, and WhatsApp using GCHQ/ NCSC- developed classifiers trained on multimodal signals like unnatural facialmicro-expressions, lip- sync desynchrony, voice spectrogram anomalies, and inconsistent lighting/ murk in vids achieving 92 delicacy in 2025 Scottish by- election tests.
Algorithms checkup for bedded watermarks( C2PA norms from Adobe/ OpenAI), probabilistic vestiges like Gaussian noise patterns in" perfect" skin textures, and contextual red flags similar as script mismatches( e.g., campaigners in unnatural crowd responses).
"Digital halves" pre-generated voice/ face models of appointees created at nomination enable forensic matching via blockchain hashes, driving 90-alternate cautions to Electoral Commission capitals. Platforms admit API feeds for automated takedowns under Online Safety Act clauses, with mortal chairpersons vindicating borders via Hive Moderation- style ensembles.