UNSW students deploy AI bots to sway a fictional referendum in 'Kingstonia'

Researchers at the University of New South Wales (UNSW) in Sydney are again running a contest in which students try to rig an election, with the twist that both the manipulators and the voters are software. In the 2026 edition of Capture the Narrative, student teams program AI bots to steer the opinions of AI-driven simulated citizens, ABC News reported on October 31.

How the game works

This year the setting is Kingstonia, a fictional island nation holding a referendum on whether people must show proof of vaccination against "dolphin flu" to enter public spaces. Teams are hired by either the yes or the no campaign, according to the competition website, which says the contest launched on October 19 and offers cash prizes.

The action takes place on Legit Social, a made-up social network built by the team of Hammond Pearce, a senior lecturer in computer science at UNSW. Bots post, like, share and reply, while the simulated citizens read that content and gradually change their views. Teams score points for capturing attention and shifting opinions, but also for spotting rivals: they can flag posts they believe come from another team's bots rather than from simulated citizens, and earn points when they are right. Pearce told the ABC the game helps students recognise manipulation and gives academics a testbed for studying AI-driven electoral interference.

What last year showed

In a January article published by UNSW, Pearce and fellow senior lecturers Alexandra Vassar and Rahat Masood described the first edition: 108 teams from 18 Australian universities, a four-week campaign, more than 7 million posts, and over 60% of the content generated by competitors' bots. The left-leaning candidate won by a very narrow margin, but when the authors re-ran the vote without interference, the right-leaning candidate won with a 1.78% swing. They concluded the bots had changed the outcome, in a controlled simulation rather than a real election. A preliminary paper on that edition, cited by the ABC, found that some teams leaned on sheer posting volume because the scoring rewarded engagement, a weakness in how such games measure success.

Why it matters

The exercise comes weeks after Australian Electoral Commissioner Jeff Pope told a parliamentary inquiry into AI that the technology could amplify existing election threats, from misinformation and foreign interference to harassment and damage to social cohesion. The simulation's limits are clear, however: simulated citizens are not real voters, and results from a game cannot show how far such tactics would work on actual platforms.