Could AI tilt the outcome of elections?

Hindustan Times

01-08-2025

THE GLOBAL wave of elections in 2024 was heralded as the first of the artificial-intelligence (AI) era. But opinions about what that meant varied. Would AI prompt a glut of malign deepfakes and large language model-generated spam? Or empower politicians to speak to constituencies segregated by language, and voters to find new sources of information to guide their choice?
With the polls closed, votes tallied, and new and returning politicians ensconced in office, the predictions of widespread AI-powered campaigning turned out to have been accurate. An exhaustive database of elections assembled by academics representing the International Panel on the Information Environment shows that AI was used in four out of five ballots last year.
The group, which models itself on the IPCC and the role it plays monitoring climate change, found AI in elections—from America to India and Britain to Indonesia. The few democracies that saw little impact from the technology tended to be small, with Botswana, Mongolia and Togo the only countries with populations larger than 2m where AI did not make an appearance.
And that impact was generally unhelpful. Most of the usage was in content creation, with the ultimate source often unknown: a little under half the time (46%), the origin of content created using AI was untraceable by the panel. Of what could be traced, almost as much came from foreign actors (20%) as from local AI-savvy political groups (25%).
There are reasons to keep one's identity secret that fall squarely within the rough and tumble of competitive politics. Using AI to dub a politician's speech into a minority language, for instance, could win respect from underrepresented groups—but might be less compelling if the video carried the logo of the management consultancy responsible. When domestic governments openly used AI tools, it was always benign, as in Madagascar, where the technology was used to help voter identification.
Harmful uses, such as when popular dead Indian politicians were re-animated for endorsements, or when a deepfake purported to show a Bangladeshi candidate dropping out of her race, made up more than two-thirds of the recorded AI usage. Foreign actors were uniformly hostile: in the Solomon Islands, for instance, the country's 750,000 citizens were targeted by Russian and Chinese actors.
One question remains unanswered. Did any of the assorted attacks, experiments and jokes actually matter? Some clearly generated discord, as in Romania, where the constitutional court noted that AI was used in an electoral-interference campaign that saw the country's presidential election annulled and re-run. But the panel's mission mirrors the IPCC's in more ways than one. The environment is changing, both groups' research confirms. But what to do about it is a decision for politicians, not panels.