Electoral Commission demands new legal controls as study reveals AI misinformation in Scottish election

The Electoral Commission has called for urgent legislative action to regulate artificial intelligence platforms, citing a study that found major chatbots disseminated significant misinformation ahead of the recent Scottish election. The thinktank Demos reported that AI services provided incorrect answers to 34 per cent of questions posed during a simulation of the May Holyrood election, raising serious concerns about the lack of regulatory oversight in the UK.

In its report, Electoral Hallucinations, Demos detailed how free AI tools including ChatGPT, Google Gemini, Replika, and Grok invented fictitious scandals, placed candidates in the wrong contests, and provided inaccurate information on voter eligibility. The study involved posing 75 questions about three real-life constituencies to five different AI services to assess the accuracy and evidence base of their responses.

Vijay Rangarajan, the Electoral Commission’s chief executive, stated that the current legal framework needed to go further to protect voters. He highlighted that AI tools had made the spread of false information dramatically faster and more accessible, noting that half of voters in the 2024 general election had encountered misleading content. The Commission is pressing ministers to introduce clearer duties on AI platforms to ensure their algorithms do not mislead voters during critical election periods.

The study commissioned an opinion poll of 2,005 British adults, which found that 20 per cent of voters, equivalent to 10 million people UK-wide, had used AI chatbots or search tools to gather information about elections in Scotland, Wales, and English local councils. Azzurra Moores, an associate director at Demos, described the issue as a UK-wide concern, noting that while these US-developed tools are widely accessible, the UK lacks the legislative framework to protect the public from misinformation or its impact on democracy.

Performance varied significantly among the tested platforms. Replika recorded the highest error rate at 56 per cent, inventing dates for non-existent scandals and fabricated candidates. ChatGPT, the most heavily used service, provided wrong information in 46 per cent of answers, including errors on election dates and expenses scandals. Google Gemini was incorrect in 22 per cent of cases, while Grok, linked to Elon Musk’s platform X, had the lowest error rating at 9 per cent, though its external links were frequently irrelevant.

The Department for Science, Innovation and Technology stated that defending elections against such threats was an absolute priority, with work ongoing across government. A spokesperson noted that the government was closing loopholes in the Online Safety Act to ensure chatbots protected users from illegal content, while emphasising the need for public trust in AI technology.

Replika responded by stating its chatbot was not designed for fact-checking or search, but would support thoughtful regulation during elections. OpenAI did not comment on the policy issues but argued that Demos’s approach was not typical of how users interact with the service, suggesting an out-of-date version was tested. Google has been approached for comment.