In the quiet spaces between elections, a new kind of battle for hearts and minds is being waged, one that most voters will never see. It is not fought with armies or bombs, but with carefully crafted algorithms, thousands of anonymous accounts, and an endless stream of content designed to plant doubt, deepen division and pull invisible strings. According to a fresh threat intelligence report by the US-based AI company Anthropic, a commercial platform allegedly used artificial intelligence to profile Malaysian voters, operate about 1,000 fake X accounts, and spread fabricated political allegations across the country. The findings were shared by Eric See To on his Facebook page, drawing public attention to a troubling dimension of modern political campaigning. Anthropic says it identified and removed an account that was using its Claude AI models to run what the company described as a commercial election-manipulation platform. Far from a random assortment of online trolls, the operation was marketed as an “influence-as-a-service” capability, and investigators linked it to Istanbul-based technology company BBS Bilisim Teknolojileri. The platform’s own documentation described the system in startling terms: a “military-grade, AI-driven, real-time political operations ecosystem.” In other words, what once would have required a state intelligence agency or a massive public relations budget can now be ordered like a software package, with the sort of scale and precision that would have been unthinkable just a few years ago. The report is a reminder that the digital battlefield is no longer a distant conflict zone; it can reach directly into a voter’s smartphone, distorting the news they read and the political conversations they join. It is tempting to shrug off such reports as alarmist, but the investigation was based on actual evidence, and the company behind it has little reason to invent a threat. The real shock is not that someone tried to build such a weapon, but that the weapon was built with tools that are already in everyday use. The report focuses on Malaysia, but its lessons are global. In an age where anyone with an internet connection can create believable content, the difference between an authentic public conversation and a manufactured one becomes harder to discern. It is also a wake-up call for social media platforms, which have become the primary arena for political debate, and for citizens who rely on them for information.
At the center of the alleged operation was a chilling ambition: to influence politics in Malaysia chart by chart, constituency by constituency. Anthropic reported that the operators ingested real census and electoral data and built a targeting system that covered all 222 parliamentary constituencies in the country. This was not a crude spray-and-pray campaign. It was a surgical, data-driven attempt to identify political and social fault lines and then exploit them. In particular, the system was allegedly designed to target sensitive issues around race, religion and royalty — topics that carry enormous emotional weight in Malaysian society. The use of such data is deeply concerning, because it suggests the people behind the platform understood not only the country’s demographics but its psychology. They knew which wounds could be reopened, which resentments could be inflamed, and which institutions could be undermined. The operation also involved a network of around 1,000 fake X accounts, a fake news outlet called Malaysia Pulse, and fabricated intelligence dossiers targeting an opposition politician and civil society organisations. The fake outlet was presumably designed to give the content the appearance of legitimacy, while the dossiers could be used to smear, intimidate or create confusion. This is a far more sophisticated operation than the usual Facebook memes or WhatsApp forwards that Malaysians may have grown accustomed to. It was built to look like an organic groundswell of opinion, even though every layer of it was manufactured. Perhaps the most unnerving part is the way the operation treated politics as a kind of engineering challenge. By feeding real census data into an algorithm, the operators could segment voters by ethnicity, religion and region, then tailor messages to suit each group. Instead of seeking common ground, the system mined existing tensions and amplified them, gambling that fear and anger are easier currencies than hope and trust. This is not just a threat to electoral outcomes; it is a threat to the very idea of a shared national conversation.
The mechanics of the operation reveal just how far AI-powered manipulation has come. According to Anthropic, the fake accounts were equipped with measures to evade detection, including regularly renewing cookies and IP addresses. That means the people behind the operation were actively working to stay one step ahead of platform security, making their activity harder to trace and dismantle. The fake accounts would likely cross-post, retweet and interact with one another to create the illusion of organic chatter, while their handlers watched their performance through custom dashboards. Anthropic’s investigation also uncovered a dashboard that allowed operators to determine how many artificial views should be generated for individual targets. In one instance, the system recorded a request for one million artificial views on the account of Malaysia’s sitting Prime Minister. To put that number in perspective, a million views can shape the trending algorithms, create the impression of widespread support or outrage, and influence how other people perceive a political figure. It can also intimidate opponents, inflate a message’s credibility, and make a false claim appear to be universally discussed. However, Anthropic was careful to stress that the engagement figures recorded by the platform were self-reported by the operators and could not be independently verified. It is possible that the system did not actually deliver all the views it claimed, or that the numbers were inflated to impress potential clients. But regardless of whether the full amount was delivered, the existence of such a request shows the mindset of the operation: a willingness to artificially distort public reality on a massive scale. The dashboard itself is a chilling artifact, turning human beings into numbers, and political influence into a metric that can be dialled up or down at will. There is something deeply dehumanising about it, because real people, with real anxieties and hopes, become the target of a statistical exercise in manipulation.
The use of AI, specifically Anthropic’s own Claude models, was not incidental to the operation. According to the report, Claude was used throughout the campaign in a variety of ways: to build custom dashboards for managing the fake-account network, track engagement metrics, develop the voter-targeting system, and support the creation and rewriting of fabricated content. This is worth pausing on. AI was not just used to write posts; it was used as the engine of the entire operation. It helped organise the fake accounts, monitor their performance, and refine the messages being pushed out. It made it possible for a relatively small group of operators to maintain a coordinated, cross-platform campaign that would once have required a large staff and substantial resources. AI also lowered the cost of producing content in multiple languages and styles, making it easier to create a fake news site like Malaysia Pulse or to rewrite a fabricated dossier so that it sounds plausible. Anthropic classified the campaign as Category Two on its Breakout Scale, meaning the assets were distributed across multiple platforms but there was no evidence they had broken into authentic communities. This is an important and somewhat reassuring nuance. It suggests that, at least in the case identified by Anthropic, the operation had not successfully infiltrated genuine political conversations or fooled large numbers of real people into adopting manipulated content. Yet the report should still sound an alarm, because the technology is improving quickly and the barriers to entry are falling. If a commercial firm in Istanbul could build something like this using publicly available AI tools, then the same playbook could be copied by actors elsewhere, perhaps with even fewer resources or less scruple. The question is not whether the operation was fully successful, but how many other operations are already out there, hidden in plain sight.
For Malaysia, the implications are serious. The country’s rich diversity is a source of strength, but it also means there are many lines of identity — race, religion, language and loyalty to the monarchy — that can be exploited by bad actors. In a society where trust in institutions is already fragile, the discovery of an AI-powered operation targeting all 222 constituencies could deepen cynicism, polarise communities and make it harder for citizens to agree on basic facts. The alleged targeting of an opposition politician and civil society organisations is especially troubling. These are the very spaces where dissent and accountability are supposed to live. If they can be smeared by fabricated dossiers and artificial online mobs, people may become afraid to speak out, and civil society may lose credibility in the eyes of the public. At the same time, the report is not evidence that the operation succeeded. Anthropic’s own classification suggests that the fake accounts and fake news outlets had not yet broken through to authentic communities. But the danger is not only in what happened; it is in what the report reveals about the future of political discourse. As AI becomes more powerful and accessible, the line between real public opinion and manufactured noise could blur even further, and the ability of ordinary people to tell the truth from a lie may be overwhelmed by the sheer volume of content. This is why digital literacy is no longer an optional skill. It is as essential to democratic life as voting itself. Malaysians need to know that such operations exist, not to become paranoid, but to become more discerning. They need to ask who is behind a viral post, whether an outrage story is being pushed by real people or by bots, and whether the information they are about to share has been verified.
The report from Anthropic should be read not as a declaration of defeat, but as a call to action. It is a reminder that as technology evolves, so do the threats to democracy. No single government, platform or company can solve this alone. Social media companies need to be more transparent about how they detect and remove coordinated inauthentic behaviour, and they need to share information with researchers and the public. Governments need to update their laws and regulations to address the new reality of AI-powered influence operations, while being careful not to let those same laws be used to silence legitimate dissent. And citizens, perhaps most importantly, need to become more critical consumers of information. Behind every algorithm are people making choices: the people who built the tool, the people who deployed it, and the people who shared the content without questioning it. The best defense is not to stop using technology, but to use it with awareness, to verify before sharing, to remain curious about who is behind a piece of content, and to find strength in real communities rather than digital mobs. If Malaysians — and people everywhere — can do that, then these efforts to manipulate will be less likely to succeed. The quiet battle will continue, but it does not have to be lost. Indeed, the very fact that Anthropic published this investigation is a promising sign. It shows that the tech industry is capable of self-examination, and that research into AI misuse can be a powerful tool in its own right. But transparency alone is not enough. There must also be accountability. Companies should be expected to report suspicious activity, to cooperate with law enforcement, and to build safeguards into their models from the start. Meanwhile, ordinary citizens should not underestimate the power of small, stubborn habits: checking a source, reading beyond the headline, refusing to share something that plays on raw hatred, and talking through political disagreements face to face. In the end, democracy is not an automated system. It is a human one — and it will survive only as long as people choose to protect it.