AI tools let a few people produce the volume of fake material that once needed an army of paid operators. In a region with young, heavily online electorates, that makes election disinformation cheaper, faster and harder to trace.
Key takeaways
- Generative AI tools reduce the time, staff and money needed to produce fake news sites, social media personas and forged documents.
- Many south-east Asian countries have young, highly connected populations, so more voters can be reached through the channels where false content spreads.
- Several countries in the region already have paid online influence operations, which could add AI tools to methods they already use.
- Producing fake content has never been the main limit on disinformation; reaching people and being believed matter more, and AI changes those less clearly.
AI removes the labour bottleneck in disinformation
The argument is simple. Running a large disinformation campaign used to take a lot of labour. The Guardian reports that such an operation once meant an army of cyber troops: people who built fake news websites, ran networks of social media accounts and forged documents to spread false stories widely. The paper reports that some experts now think AI alone can do much of that work, and that it could be used as a weapon in south-east Asia.
The claim is not that AI invents a new kind of threat. Fabricated stories, fake accounts and forged papers are old tactics. The claim is that AI changes the cost of using them. Whatever once needed dozens or hundreds of people writing posts, designing websites and keeping fake identities going can now be done, at least partly, by software that produces text, images, audio and video on request.
Lower costs matter in three ways. More actors can afford to run a campaign, including small political groups, local interests and individuals who could never pay for a troll farm. Campaigns can run faster, reacting to news within hours. And campaigns become harder to trace to their source, because fewer people are involved and fewer payments or organisational links are left behind.
The region makes these effects more serious. The Guardian notes that south-east Asia is home to countries whose populations are mostly young and highly connected. Elections there are fought more and more on social media and messaging apps, where content spreads quickly and is checked slowly. Where those conditions meet cheap AI-generated material, the risk to how fair elections are and how far people trust them is likely to rise.
Generative tools collapse the cost of producing fake content
The first piece of evidence is what current generative AI systems can openly do. Large language models write fluent text in many languages, in many styles and voices, in seconds. Image generators produce realistic pictures of events that never took place. Voice-cloning and video-synthesis tools can make recordings that appear to show real people saying things they never said. Material of this kind is usually called a deepfake.
Each of these tools replaces a job that once needed human labour. A fake news website needs a steady flow of articles, and a language model can write them. A network of fake social media accounts needs profile pictures, life stories and a believable posting history, and generative tools can supply all three. A forged dossier needs convincing documents, and AI systems can imitate official formats and wording. This is the change the Guardian describes: from a large coordinated operation to something much smaller.
Language is a key part of this. South-east Asia is linguistically varied, and influence campaigns once needed native speakers to write content that read naturally in each local language. Multilingual AI models lower that barrier, although how well they handle the region’s less widely used languages and dialects varies and is not fully documented.
Scale matters as well as cost. Generated content can be produced in large volumes and in many versions. That can make platform detection harder, because systems that look for repeated or copied material may not flag many slightly different posts. It can also flood an information space so that accurate reporting is harder to find.
Young, connected populations widen the reach
The second piece of evidence concerns the audience. The Guardian describes the populations of many south-east Asian nations as mostly young and highly connected. The source gives no specific figures, so this article does not cite any. The general pattern is still relevant to the argument.
When a large share of voters get political information mainly through social media, short-video platforms and messaging apps, the routes by which disinformation spreads become central to public debate. Those platforms reward content that provokes strong reactions. They spread material through personal networks, where people tend to trust what friends and family share. Encrypted messaging apps add a further problem: content passed around in private groups is largely invisible to fact-checkers, researchers and the platforms’ own moderation systems.
Being young and connected does not make voters more gullible. Younger users are often more familiar with digital tools and with the idea that online content may be fake. The point is about reach and speed. An electorate that lives online can be reached cheaply and quickly by anyone who can produce content to suit those channels, and generative AI is good at producing exactly that kind of content: short videos, eye-catching images and posts written in casual everyday language.
Elections add to the pressure. Campaigns are short, and false claims released close to polling day may not be corrected in time. Synthetic audio or video of a candidate, released just before a vote, is the kind of tactic that rapid AI generation makes easier and that a highly connected electorate would see quickly.
The region already has an industry of paid online influence
The third piece of evidence is that the infrastructure for disinformation already exists in parts of south-east Asia. The Guardian’s reference to armies of cyber troops points to a familiar feature of politics in the region. In several countries, researchers, journalists and civil society groups have documented paid networks of online operators who push political messages, attack opponents and promote favoured narratives. These are often hired through public relations firms, marketing agencies or informal intermediaries.
This matters because AI tools are most dangerous when they are added to an operation that already works. A network that already knows how to build an audience, game platform algorithms and time its messages for effect can use generative tools to do more, more cheaply. The people who once wrote the content can move into strategy, distribution and managing networks, while software takes over production.
Having such networks also shows there is real demand for online influence services during elections. Where political actors are already willing to pay for manipulation, any tool that lowers costs or raises output is likely to be adopted. How far AI has already been built into these operations, and on what scale, is not established in the source material and is not known from it.
The strongest case against the argument
The strongest objection is that producing content has never been the main limit on disinformation. Even before generative AI, fake material was cheap to make. A forged screenshot, a misleading caption on a genuine photograph or a clip taken out of context costs almost nothing. Much of the most effective political disinformation has been crude rather than sophisticated. If cheap content was already plentiful, making it cheaper may change less than it seems.
On this view, the hard parts of a disinformation campaign are distribution and credibility. Content has to reach an audience, get past platform defences and be believed. That depends on trusted accounts, established networks, existing grievances and the willingness of real people to share it. AI does not obviously create trust or audiences. A flood of generated posts from new, unknown accounts may be ignored or caught more easily than a few well-placed messages from influential figures.
A second objection is that the defences are also changing. Platforms, election authorities, fact-checking groups and researchers can use AI to detect coordinated behaviour, flag synthetic media and analyse content in many languages at scale. Some AI developers add safeguards or markers meant to identify generated content, though how effective these are is debated. The balance between attack and defence is not settled in favour of the attackers.
A third objection is about evidence. Warnings that AI will transform election manipulation have been common, but showing that a given campaign changed how people voted is very hard. Persuasion through media in general tends to have limited measurable effects on firmly held political views. If AI-generated disinformation mainly reaches people who already agree with it, its effect on results could be small, even if its effect on the general information environment is real.
These objections do not show that the risk is imaginary. They suggest the size of the risk may be smaller, or different, from what the most alarmed accounts assume.
Evidence that would change the conclusion
Several kinds of evidence would show whether AI really is making election disinformation in south-east Asia worse, or whether the concern is overstated.
Systematic studies of recent elections in the region would be the most useful. If researchers found that AI-generated content made up a growing share of the false material that circulated, and that it reached large audiences, the argument would be stronger. If they found that most widely shared disinformation was still made by hand, crude and spread through existing influential accounts, the argument that AI is a decisive accelerant would be weaker.
Evidence on effects would also matter. Studies showing that exposure to synthetic media changed how voters saw candidates, lowered turnout or damaged trust in results would support the case for serious concern. Studies showing little measurable effect on behaviour would point the other way.
Evidence on detection matters as well. If platforms and independent monitors can reliably find and limit AI-generated election content, including in local languages and on messaging apps, the lower cost of producing it would matter less. If detection keeps failing, especially in private channels, the risk would look greater.
Finally, documented links between AI tools and the paid influence networks already working in the region would be telling. Clear evidence that those operations had cut staff, grown their output or reached new audiences by using generative AI would directly support the central claim. Without such evidence, the argument rests mainly on reasoning about costs and capabilities, which is plausible but not yet shown.
Sources and further reading
- The Guardian: technology reporting on concerns that AI could increase election-related disinformation risks in south-east Asia.
- Academic research on networked disinformation and paid online influence operations in south-east Asian politics.
- Transparency and threat reports published by major social media platforms on coordinated inauthentic behaviour.
- Reports by independent fact-checking organisations and election-monitoring groups working in the region.
Surfaced from the rss:guardian_tech signal “AI election disinformation risks”. AI-assisted draft, editorially reviewed.
