The release of synthetic election videos and advertising by One Nation and several lobby groups in the lead-up to Victoria's November state poll has signalled a new chapter in the use of AI by mainstream political actors in Australia.
But it's nothing compared to what's happening in the US, where candidates and political action groups have already spent more than US$137 million this year on promoting AI-generated TV and online ads ahead of the congressional midterms.
That figure, compiled by the Wesleyan Media Project, compares to less than US$50,000 in 2024 and is almost certainly an undercount.
Ads were only counted if AI use was reported in the media, obvious to researchers or explicitly disclosed. Only 30 of 50 US states require AI labels during election campaigns, and these were present in just 27 per cent of the 220 ads identified.
The data reveals an interesting political split, with 81 per cent of AI ads paid for by Republicans, 12 per cent by Democrats and the rest by other sponsors.
It's early days, but in Australia the use of AI also appears to be more common among right-wing political operatives, notwithstanding a low-quality deepfake of a dancing Peter Dutton shared by federal Labor's TikTok account in 2024.
During the current Victorian campaign, a new lobby group named Fix Victoria spent around $220,000 in less than two months promoting AI-generated ads on Facebook and Instagram, data from Meta's Ad Library shows. The Guardian found the group's senior ranks include several former staff from conservative think tank the Institute of Public Affairs.
After One Nation launched its Victorian campaign with a controversial synthetic video laden with stereotypes, conservative lobby group Advance released its own AI video depicting a "nightmare version of election night" in which current and former Labor premiers celebrated wildly after voters "forgot to put Labor last".
While not the first time Advance has produced a campaign video starring a deepfake of a real political opponent, it's arguably the most realistic effort in Australia to date, even if it reportedly went down badly with the state Coalition.
In the US, these types of deepfakes are becoming commonplace. The New York Times has presented a swathe of examples from the Wesleyan dataset, reporting that nearly half of the total "contained AI-generated depictions of real politicians".
Eryn Newman, a professor of psychology at the University of Waikato, told The Repost that attention-grabbing AI imagery had the potential to increase the amount of fictional or misleading content that people consume.
"The problem is that we are consuming vast amounts of information in our digital environments and there is a risk that we remember the gist of the message, but not the fact that it was AI or completely fictional," she said. This could have "more insidious effects on how we understand information and our associated beliefs", Professor Newman added.
Many of the scenes depicted in the US ads — and the Advance video — may be too absurd to be believed. But with AI, the line between satire and deception is easily blurred, as real politicians can be made to not only look very foolish but also say things they never said.
Professor Newman said research was still emerging on the extent to which obvious deepfakes affect how audiences subsequently perceive the real person or correctly remember who said or did what.
As discussed in our article on AI labelling, one recent experiment found that people shown a deepfake video of someone confessing to a crime or transgression still relied on that video to form opinions of the real person's guilt — even after being warned the video was an AI fiction.
Experts have also told The Repost that the purpose of deepfakes is often not to deceive but to encourage sharing and to entrench pre-existing beliefs.
In recent years, South Australia and NSW have both introduced laws prohibiting deepfakes of real people in ads and other political materials during state election campaigns. But beyond narrow rules against misleading voters about how to vote, Australia has no national laws that require truth in political advertising or the labelling of AI-generated content.
This means regulating and labelling deepfakes in Australian political ads largely falls to broadcasters and online platforms. On social media, rules vary substantially across platforms, with a new report from the Institute for Strategic Dialogue identifying "meaningful gaps" in their approaches.
