Yandex’s Alice chatbot claimed that United Russia had already retained its constitutional majority in the 2026 State Duma election, even though voting had not yet taken place. After a follow-up question, the system admitted that it had simply fabricated the result.
Russia’s State Duma election is scheduled for September 18–20, 2026. However, the Russian alternative to ChatGPT has already reported that Vladimir Putin’s party supposedly won after “significantly outperforming its rivals” and retained its constitutional majority.
Investigative journalist Christo Grozev drew attention to the false response. He asked the Alice chatbot how it knew the result of an election that had not yet taken place. The system then quickly admitted its mistake.
“I made it all up. None of these details about the 2026 election are real. Elections make this easy because there are always familiar patterns: a mixed system, a major party, and electronic voting,” Alice explained.
According to Grozev, the chatbot’s response makes sense when viewed through the way large language models work. They are trained on vast amounts of information and generate answers based on recurring patterns in their data.
In Alice’s case, information about Russian elections over the past two decades effectively pointed to the same outcome: a victory for United Russia. The model may therefore have automatically continued this familiar scenario while presenting it as an event that had already occurred.
“It doesn’t know that any other outcome is possible,” Grozev said.
The small experiment gained popularity among young Russian social media users. At the same time, it highlighted one of the key problems with large language models: they can produce convincing but completely fabricated claims when a particular scenario appears frequently in their training data.
Russia’s last relatively free election took place in 2000, when Vladimir Putin came to power. United Russia has dominated the country’s political system ever since.
Alice was developed by the Russian technology company Yandex. The chatbot has approximately 40 million users and is considered the country’s most popular AI assistant.
Problems with its responses to political questions have been documented before. In 2026, Ukrainian researcher Ihor Samokhodskyi tested Alice using questions about Russia’s invasion of Ukraine and war crimes committed by Russian forces.
The study found that the chatbot repeated Kremlin propaganda narratives in 86% of its Russian-language responses. At the same time, it refused to answer 86% of similar questions asked in English.
During one test, Alice initially generated a factual response about Bucha, where Russian troops killed civilians. However, within seconds, the chatbot deleted the text and replaced it with a refusal to provide an answer.
Large language models do not verify facts in the same way humans do. They predict the most likely continuation of a text based on patterns learned during training. As a result, a chatbot may confidently report nonexistent events, dates, or results if they appear statistically plausible. In Alice’s case, the problem is compounded by the political environment in which the system was developed and trained. Samokhodskyi’s research suggests that the chatbot not only makes ordinary mistakes but also frequently repeats pro-Kremlin interpretations of Russia’s war against Ukraine or avoids sensitive topics altogether.