AI Chatbots Spread Misinformation to Voters Before Scottish Elections

A study by the independent analytical center Demos has revealed critical problems with information quality in popular AI chatbots. Before recent Scottish elections, ChatGPT, Gemini, Grok, Replika, and other services provided voters with misinformation in a third of cases, inventing fictional scandals and fabricating non-existent candidates. The results raised concerns among regulators and cybersecurity experts.

What Demos Found

Demos researchers conducted systematic testing — they asked AI services over fifty questions about the upcoming Scottish elections, voting procedures, candidates, and voting results. The results proved alarming and unexpected. Overall, 34% of all responses contained misinformation.

The AI services didn't just make mistakes in details — they created entire fabricated narratives. Some chatbots invented political scandals that never existed in reality. Others fabricated candidates who didn't participate in the elections.

Still others provided completely incorrect voting dates. The most dangerous aspect of these errors is that AI delivered all this with complete confidence, without caveats about uncertainty or acknowledgment of its limitations. A user reading such a response could completely believe the fabricated fact and make an important decision based on it in the voting booth.

Demand for New Rules

The Electoral Commission, Britain's election watchdog, issued an official demand following publication of the research results: introduce new legal mechanisms for controlling AI platforms. The Commission emphasized a critical problem: the current legal framework of Great Britain does not provide sufficient tools to combat misinformation distributed by AI systems.

"We see a real problem: AI services are becoming increasingly influential sources of information for citizens, but at the same time remain practically without any oversight," the

Electoral Commission stated.

According to regulators, the following measures need to be implemented:

• Mandatory labels and warnings on AI responses about the risk of misinformation
• Technical requirements for providers: clearly document the model training date and specify the limits of its knowledge
• Legal liability of platforms for spreading false information during elections
• Complete transparency in model training — disclosure of what data was used

Why This Is Serious

The problem becomes increasingly acute as AI popularity grows. ChatGPT, Gemini, and similar systems are already used billions of times per month by people searching for information. Yet these models remain unreliable on questions requiring current information about specific events. AI is trained on data up to a certain date, so its knowledge inevitably becomes outdated. When the system doesn't know the answer, instead of simply admitting ignorance, it often "hallucinates" — inventing plausible facts that sound convincing to the average user. For elections and other critical information, this represents a real threat to the democratic process.

What This Means

The Demos study confirmed what information security specialists have long warned about: AI chatbots are dangerous in the context of political elections and important public decisions. Regulating AI platforms becomes not just a matter of technological responsibility, but also a matter of protecting democratic processes. For regular users, the conclusion is straightforward: don't rely on AI as your primary source of information about politics, elections, or other critically important events. Always verify information through official sources, authoritative news publications, and government resources.