It began with a face that many Indonesians knew and trusted. In August 2025, a video clip began racing through social media feeds, showing the then-Minister of Finance, Sri Mulyani, speaking before a crowd. Her voice was clear, her gestures familiar, and the words were shocking: “Teachers are a burden to the state.” Within hours, the clip had sparked outrage, disbelief, and heated arguments across platforms. Many people took it at face value, convinced that a senior official had really said something so dismissive of educators. But the video was a deepfake, a highly realistic fake created with artificial intelligence. Deepfake technology uses machine learning to manipulate or completely fabricate images, audio, and video, making it possible to put words into someone’s mouth or make them appear to do things they never did. What made the Sri Mulyani case so alarming was not just that the video looked authentic, but that it was so easy to make. Generative AI, the type of artificial intelligence that can produce new content from simple prompts, has become widely accessible in recent years. Tools that once required technical expertise are now available to anyone with a smartphone, and they are increasingly capable of creating media that is nearly impossible to distinguish from reality. The Sri Mulyani clip was only one example, but it became a powerful symbol of a much larger problem: the rise of hoaxes manufactured by AI. As the technology improves, the line between fact and fabrication blurs further, and ordinary people are being forced to question whether what they see and hear online ever really happened. For Indonesia, a country that has long battled fake news and political disinformation, this new wave of AI-generated deception has become a growing threat, and fact-checkers are scrambling to understand its scale and prepare for what comes next.
To get a clearer picture of just how quickly AI-generated hoaxes have spread, the Indonesian news organization Kompas conducted an extensive review of articles on Turnbackhoax.id, one of the country’s leading fact-checking platforms. This website is owned by the Indonesian Anti-Defamation Society, or Mafindo, a civil society group dedicated to debunking false claims. With permission from Mafindo, Kompas accessed thousands of articles through an application programming interface, or API, and downloaded everything published between 2022 and August 2026. In total, the research team pulled 10,212 articles containing examinations and refutations of hoaxes that had circulated in Indonesia. From there, they filtered the data using tags, keywords, and a large language model to identify which hoaxes were suspected of being created by AI. The results were striking. Of the more than ten thousand articles, 665 dealt specifically with AI-generated hoaxes. The yearly breakdown revealed a clear trend: in 2022, only 2 such cases were found; in 2023, the number rose to 19; then in 2024, it jumped dramatically to 239. In 2025, there were 233 cases, and in just the first eight months of 2026, there were already 172. In absolute numbers, 2024 saw the most AI hoaxes, a spike that was likely connected to the general elections held that year, when political tensions and misinformation typically surge. But looking at proportions, the trend becomes even more concerning. In 2024, AI-generated hoaxes represented only about 6.46 percent of all hoaxes examined that year. In 2025, that share rose to 15.77 percent, and by August 2026, it had climbed again to 17.39 percent. In other words, while the total number of hoaxes may fluctuate depending on political events, the role of AI in creating those hoaxes is growing steadily and showing no signs of slowing down.
The subjects of these AI-generated hoaxes reveal a pattern that is both predictable and troubling. Politics dominates the list by a wide margin. Out of the 665 hoaxes analyzed, 212 were political in nature. This cluster includes false stories about international conflicts such as the Iran-United States-Israel tensions, the Israel-Palestine conflict, and the situation in Papua, as well as domestic issues like the 2024 elections, corruption and misuse of state funds, and protests and riots that took place in August and September 2025. The heavy political focus is hardly surprising, given how deeply Indonesians engage with political news and how easily divisive content spreads online. After politics, the next most common category was aid-related hoaxes, with 85 cases. Many of these involved supposed distributions of cash assistance, social assistance, or educational funds, often accompanied by forged announcements or fabricated statements from officials. Business-related hoaxes followed with 73 cases, and here the pattern was largely about impersonating public figures. Fake videos and images were used to promote online gambling, health products, loans, and investments, often making it appear as though respected business leaders or celebrities were endorsing these schemes. A similar misuse occurred in the theme of prizes, where 71 hoaxes exploited public figures in connection with fake giveaways of cash. Across all these categories, one thing stands out: AI is being used not only to spread false political narratives but also to commit fraud and deceive ordinary citizens who are hoping for financial help or a better life. The people whose faces and voices are stolen in these hoaxes often have no idea their identities are being used, and by the time the truth comes out, the false content has already traveled far and wide, reaching thousands, sometimes millions, of viewers.
The way AI-generated hoaxes spread online says a great deal about how Indonesians consume information. When the researchers examined the platforms where these hoaxes were shared, Facebook was by far the most dominant, accounting for 389 pieces of content out of the 665 analyzed. TikTok followed with 175, while X, formerly known as Twitter, had 44, Instagram had 38, YouTube had 17, and Threads had 9. It is worth noting that many hoaxes were shared across multiple platforms at once, multiplying their reach and making them harder to contain. Perhaps even more important than the platform breakdown is the format of the content. Video dominates overwhelmingly, representing 560 pieces of content, or about 84 percent of all AI-generated hoaxes in the dataset. Images made up another 102 pieces, or 15 percent, while pure text-based AI hoaxes were extremely rare, only 3 pieces of content, less than 1 percent. This makes sense in a deeply visual media culture where moving images carry an authority that text alone often lacks. A video of a politician saying something outrageous is far more believable and far more shareable than a paragraph of rumors. Bentang Febrylian, a senior fact-checker at Mafindo, explained that the surge in AI hoaxes began in 2024 when generative AI tools first became widely known to the public. In the years since, he said, these tools have improved rapidly. AI-generated videos, photos, and sounds have become increasingly smooth and realistic, and that technical sophistication is directly tied to the growing number of hoaxes that fool people. Even trained fact-checkers now need more time and better tools to determine whether a clip is authentic, and the average social media user, scrolling quickly and reacting emotionally, is even more vulnerable.
Experts who study misinformation and AI agree that the current situation is only the beginning. Mark Ufie, the chairman of Mafindo’s central board, pointed out that while the current number of AI-produced hoaxes might not yet seem overwhelming, that is largely because no major moment has triggered a massive explosion of them. But he warned that this could change quickly. The 2029 elections, in particular, are a source of deep concern. In every previous election cycle, Indonesia has experienced a sharp increase in hoaxes, and with AI tools now widely available, the next democratic festival could be flooded with sophisticated fake content designed to manipulate voters. Because of this, Mark argued that the government must act immediately, not wait until the crisis is upon them. He called for concrete regulations that address the many dimensions of AI, including its use in creating disinformation. He also suggested that Indonesia should establish a national council bringing together government, civil society, technology companies, and other stakeholders to oversee AI governance. In addition to regulation, he stressed the importance of digital literacy, saying that citizens need to be trained to question what they see online rather than accepting it at face value. He specifically proposed that digital literacy be integrated systematically into the curriculum of high schools and universities, so that young Indonesians grow up with the critical thinking skills needed to resist manipulation. Novi Kurnia, a communication studies lecturer at Gadjah Mada University, echoed these concerns but also offered a more balanced view. She reminded people that AI is not inherently evil; it is a tool that can be used for both good and bad. On the one hand, it makes it easier to create misinformation, disinformation, and deepfakes. On the other hand, AI can also be used to detect hoaxes, verify authentic content, and support fact-checking efforts. The real problem, she noted, is that many ordinary people simply cannot recognize AI-generated content when they see it. That is why broader digital literacy efforts are so essential. She also argued that social media platforms must be pressured to detect AI-generated content and label it clearly, so users understand when they are looking at synthetic media. At the same time, regulations around harmful content and AI need to be strengthened, even though she acknowledged that this is not easy. As she put it, AI has grown more diverse so quickly that regulation struggles to keep pace, and that is the fundamental challenge facing policymakers.
The Indonesian government has begun to respond, in part by preparing two presidential regulations related to artificial intelligence. The first regulation focuses on AI ethics, while the second concerns the National AI Roadmap for 2026 to 2029. Deputy Minister of Communication and Digital Nezar Patria described these instruments as important for building a responsible AI ecosystem in Indonesia. The roadmap, he explained, is deliberately designed for a five-year period because AI technology is evolving so rapidly that a longer plan would become outdated before it could be implemented. He acknowledged that the development of AI has been surprising and fast, and that regulations must remain flexible enough to adapt to new challenges without falling behind. The work of building responsible AI governance is not just about stopping hoaxes, however. It is about shaping how this powerful technology is used for years to come. AI has the potential to bring enormous benefits to Indonesia, from improving public services to driving economic growth. But without proper oversight, it also has the potential to erode trust, deepen social divisions, and threaten democratic processes. The deepfake of Sri Mulyani was a wake-up call, a sign that the future of information is already here and that no one is safe from having their words or image twisted by machines. The challenge now is to prepare a society that understands what AI can do, a regulatory system that can keep up with rapid change, and a culture of digital literacy that treats every video, image, and article with healthy skepticism. Only by doing so can Indonesia harness the power of AI while protecting its citizens from its darker effects.