Imagine scrolling through Facebook and seeing a video of a well-known politician saying something outrageous. It looks real, it sounds real, and you might share it before thinking twice. That is exactly the situation new research says has been manufactured at scale for dozens of Australian politicians. Reset Tech, an independent research organisation, has tracked more than 200 Facebook pages over the past eight months that impersonated political figures including Prime Minister Anthony Albanese, Liberal MP Angus Taylor, and One Nation leader Pauline Hanson. The pages were traced back to an influencer business based in Sri Lanka, and researchers believe what they found is only a fraction of a much larger problem. What made this operation different from traditional state-run propaganda was its commercial structure: around 10 per cent of the pages had been monetised through Facebook’s own content-payment scheme, meaning the people behind them could earn money every time their fake posts generated engagement. “This isn’t foreign interference as we traditionally think of it,” said Rys Farthing, Reset Tech’s chief research officer. “What we see here is foreign influence for commercial motivations.” In other words, the goal was not necessarily to swing an election in a particular direction, but to profit from the outrage, division and clicks that political disinformation reliably produces. This blending of political messaging and financial incentive makes the problem especially slippery, because the actors are neither straightforward foreign agents nor ordinary domestic political operatives; they are internet entrepreneurs using deepfakes as a product.
The operation appears to have been carefully designed to build a audience before the politics kicked in. Many of the pages carried patriotic, harmless-sounding names such as “I Love Australia”, “Life in Australia” or “Beautiful Australia”, and initially posted content with broad appeal: AI-generated pictures of animals, attractive women, or admiring posts about Australian landscapes and culture. This gave the pages a veneer of authenticity and helped them accumulate followers. Then, once a page had an audience, it began to pivot toward political material—or, in many cases, was political from the very beginning. The most common format was a short clip on Facebook Reels in which a politician appeared to say something inflammatory. The themes were consistent: anti-immigrant, anti-Muslim and nationalistic, often framed around controversial proposals or social tensions. Many posts used polls to bait users into interacting, asking questions like “Should this be allowed?” or “Do you agree?” According to the report, another recurring pattern involved sexualised images and videos of female politicians, used both to degrade the women and to attract clicks. Farthing said the content was not driven by a coherent political ideology but was ruthlessly opportunistic. “Sometimes it is directly related to news events, but basically … it’s just throwing spaghetti at the wall,” she said. “They work out which [posts] get them the clicks, which ones bring the money in, and then they create more like that again.” The result is a low-cost, high-volume content machine that tests what works and doubles down on it.
The practical effects can be seen in the specific cases highlighted by the researchers. Senator Fatima Payman, a Muslim who left the Labor Party to sit on the crossbench, was impersonated by a page called “Australia Info Live”. A fake AI-generated clip showed her saying, “Please don’t deport me back to Afghanistan. It’s not safe for me there. Vote no to support me.” The post drew hundreds of likes and comments and carried no visible label telling viewers it was fake or AI-generated. For someone scrolling through their feed, it could easily have been mistaken for a genuine statement, especially at a time of heated public debate about migration and national identity. Independent MP Allegra Spender was also targeted. Senator David Pocock was impersonated by a page called “Australia Life & News” in a clip that falsely claimed he had “joined forces” with One Nation. Pocock said he had not seen that exact clip, but his team had been sent “a bunch of deepfakes” of him over the previous months. He said the experience was disturbing because a well-made deepfake can be almost impossible to distinguish from a genuine video. “As humans, when we see someone say something, we generally believe that it’s them saying it,” he told the ABC. That, he argued, is why deepfakes pose such a direct threat to democracy: they erode the basic trust that voters place in the words and actions of public people.
Although the initial focus was on Australia, the network was global in ambition and reach. Reset Tech found the same Sri Lanka-based operation was also targeting France, Italy, the UK, the US, Canada and Mexico, using essentially the same playbook: innocuous posts to build an audience, then a shift to divisive political content with AI-generated footage and images. The researchers identified 1,864 coordinated Facebook pages around the world, many with almost identical branding and a combined following of 5.9 million users. Farthing stressed that this was not a handful of rogue actors but an industrialised form of what she called “AI slopaganda”. “We really wanted to highlight that this wasn’t just a unique issue with one bad actor or a couple of bad actors misusing a platform system,” she said. “Rather, the research showed networks targeting political systems on a massive industrial scale.” The report also judged Meta’s enforcement against its own rules to be inadequate. Only about one in three political AI-slop posts received an AI label, and only one in 33 was removed. Reset Tech said the Australian arm of the network violated Facebook’s policies on inauthentic behaviour, hate speech and AI-generated content. Farthing was blunt about the gap between the resources of a small research team and those of one of the world’s largest companies: “We have a handful of researchers, and we were able to find this and detect all of these networks with ease. If we can do it, I don’t quite understand how these companies that have billion-dollar revenues can’t do it.”
The problem has become urgent enough that lawmakers are now starting to respond. New anti-deepfake laws have been used for the first time, with a case involving non-consensual sexually explicit material going before the courts. Authorities and experts say the creation of such material is rising, and it is the darker edge of the same technology that produced the political videos: deepfakes can strip a person’s dignity, damage their reputation and, in a political context, seriously harm their standing before an election. Home Affairs Minister Tony Burke, who was himself deepfaked by the network, said the spread of realistic fake content had changed the way Australians need to consume information. “Last election for the first time I saw photos of myself online with people I had never met,” he said. “In more recent times, I have seen videos … and interviews of some of my colleagues saying things they have never said.” Farthing described the campaign in Australia as “foreign distortion of Australian political discourse”, adding that “there’s nothing domestic about this”. The same pressure is being felt outside politics: regulators are removing record numbers of fake celebrity-endorsed investment scams that use AI-generated images of famous people to trick consumers. The underlying pattern is identical—synthesised likenesses deployed at scale for commercial gain, whether the product is a fraudulent investment scheme or a false political attack.
Who is ultimately responsible for all this? Farthing points squarely at the platforms. “The platforms have made all of the decisions that enable their monetisation system to work, right? And this is the result.” In other words, the same algorithms that reward outrage, clicks and long screen time also reward the fake pages that manufacture outrage. Senator Pocock agreed, saying it was increasingly clear that tackling the problem was not in the interests of social media companies. “Their business model is based on outrage, on eyeballs on the screens for as long as possible, and on clicks,” he said. Meta, Facebook’s parent company, responded to the report by saying it was “reviewing the pages identified and will take action against any that violate our policies.” But critics say that reactive approach is unlikely to solve a problem already being produced on an industrial scale. The report suggests that page operators are constantly testing new formats, new platforms and new monetisation methods, and that they are several steps ahead of the moderation systems. For ordinary users, the safest approach is scepticism: political videos circulating on social media may look real, but they are increasingly cheap and easy to fake. The researchers’ warning is that the network they uncovered is probably just one example of a much larger ecosystem of AI-generated political influence. Unless platform incentives change, or regulation forces them to change, the problem will only grow. The burden, for now, is on voters to look more carefully before they believe, share, or click.