In a world where artificial intelligence is woven into everyday life, one of the most unsettling questions is how easily the same tools that help us write, learn, and communicate can be turned against the foundations of democracy. A recent case involving Malaysia offers a stark and sobering example. Anthropic, the American AI company behind the Claude models, has revealed that it disrupted a commercial election-manipulation platform that primarily targeted people in Malaysia. According to a threat intelligence report published on Thursday, Sep 10, the operation involved around 1,000 fake social media accounts, a bogus news website, and fabricated intelligence dossiers. At its core, the platform used Anthropic’s own AI systems to build a method of profiling and targeting Malaysian voters by constituency. That is a deeply personal form of manipulation, one that allows political messages to be tailored to the specific fears, hopes, and frustrations of a community. What makes this case especially troubling is that it was not a lone hacker or a shadowy state agency. It was a paid service, an “influence-as-a-service” operation, built on top of legitimate AI technology and offered to clients as a product. Anthropic said it had disrupted the activity between December 2025 and August 2026, but it did not identify who paid for the Malaysia-focused operation, nor did it say whether the work was linked to any particular Malaysian election. The company also did not specify exactly when the operation took place or how long it ran. Even with those unanswered questions, the scale and sophistication of the scheme paint a clear picture: AI-powered election interference is no longer a distant hypothetical. It is here, and it is being sold to the highest bidder.
To understand how this operation worked, imagine a political consultancy that has never met its clients, knows nothing about the country, and yet can produce a campaign that speaks directly to individual voters. The platform behind this scheme first presented itself as a defensive cyber intelligence and counter-disinformation tool. That cover story gave it a layer of legitimacy while it quietly assembled the opposite capability. Behind the scenes, it used Claude AI models to build a system that could identify the concerns of voters in different constituencies and generate content designed to influence them. It did not stop at writing posts. The operators set up about 1,000 fake social media accounts to spread the messages, making it appear as though ordinary citizens were sharing their own thoughts. To make the scheme even more convincing, they launched a fake news site and prepared fabricated intelligence dossiers, likely intended to damage or elevate certain individuals. Anthropic did not reveal the contents of those dossiers, but the pattern is familiar: manufacture a narrative, feed it to the public through a network of fake voices, and let the algorithms do the rest. The real innovation here was not the propaganda itself but the efficiency. Rather than relying on human workers to write hundreds of different messages, the platform used AI to automate the profiling, drafting, and posting. That is the new face of election interference, and it is far harder to spot than the old one.
The company behind the platform was identified by Anthropic as BBS Bilisim Teknolojileri, an Istanbul-based technology company. According to the report, there were clear links between the platform and this firm, which sold access to the system as a paid service. In other words, election manipulation is no longer the exclusive domain of intelligence agencies or ideological activists. It has become a business model, a product that can be purchased like any other software subscription. Anthropic described it as an “influence-as-a-service” capability, a phrase that sounds almost mundane but describes something deeply corrosive. A client could hire a team to interfere with the democratic process in another country without needing any technical expertise. The involvement of a Turkish company does not mean that Turkey was behind the Malaysia operation; the company may have been serving clients from anywhere in the world. Anthropic did not name those clients, nor did it say whether the operation was connected to a particular election. What is clear is that the infrastructure was built to be reusable, scalable, and commercial. That is a worrying sign for countries that may not expect to be targeted by sophisticated digital manipulation. Smaller nations, in particular, often lack the resources to detect such operations on their own, and by the time they realize what is happening, the damage may already be done.
For Malaysia, this news arrives at a moment when concerns about disinformation and foreign interference are already high. The country has a vibrant democracy and a history of competitive elections, and like many nations, it has struggled to respond to fake news. The idea that AI could be used to profile voters by constituency is especially troubling because it means manipulation can be surgical rather than broad. Instead of spreading a few clumsy lies, an operator could study a community’s specific anxieties, then craft messages that feel local, personal, and trustworthy. That kind of targeting is difficult for ordinary voters to detect and difficult for regulators to track. The bogus news site and fabricated dossiers could be used to push narratives that would be hard to debunk, especially if they were seeded through thousands of fake accounts. Malaysian authorities have not yet commented publicly. CNA contacted Malaysia’s Communications Minister Fahmi Fadzil and the Malaysian Communications and Multimedia Commission (MCMC), the national regulator for the communications and multimedia industry, which falls under the Communications Ministry. They were asked whether Malaysian authorities are investigating the operation and whether BBS Bilisim Teknolojileri or its representatives pursued any contract with the government. At the time of writing, no answers were available. The silence is itself a reminder that regulators often learn about these operations only after private companies expose them.
Anthropic’s response was to remove the Claude account involved and to use the information gathered from that account to map the wider operation. In its case study titled “Disrupting a commercial election-manipulation platform targeting Malaysia,” the company described how one suspicious account led to a web of associated accounts, domains, and infrastructure. This is a common approach in threat intelligence: a single small clue can unravel an entire network. The report covered activity Anthropic said it disrupted between December 2025 and August 2026, including cyber operations, influence operations, surveillance, scams, fraud, and other forms of misuse. The Malaysia case was just one part of that broader pattern, but it stood out because it showed how commercial actors are now using AI for political purposes. For AI companies, this is the central paradox. The same models that help people draft emails, analyze data, and create art can also be used to build powerful influence operations. Anthropic has policies against such abuse, but enforcing those policies is an endless game of cat and mouse. Every time a bad actor is removed, another one appears, often with new tricks. The decision to publish a detailed report is itself a form of defense, because it helps researchers, governments, and the public understand the threat. But it also highlights the limits of voluntary self-regulation. AI companies cannot be expected to police every user, and they cannot always predict how their models will be used in contexts they never imagined.
What should we take away from this? First, the threat of AI-powered disinformation is not hypothetical. It is happening now, in countries like Malaysia, and it is being sold as a service. Second, the defenders are often private companies, not governments. That is both a strength and a weakness. A company like Anthropic has the expertise and the data to spot misuse, but it does not have the legal authority to prosecute criminals or protect voters. That responsibility rests with governments, and they need to move faster. Third, the public has a role to play. When we see suspicious content online, especially during election periods, we should pause and question its source. The fake accounts in this operation were designed to look like real people, but patterns often emerge: unusual posting times, identical language, or a sudden flood of coordinated messages. Finally, this case is a reminder that technology is never neutral. AI can be a force for education, creativity, and connection, but it can also be used to erode trust, distort reality, and undermine democracy. The best defense is not to fear the technology but to understand it, to demand transparency from AI companies, and to support independent investigations. The election-manipulation platform in Malaysia was disrupted, but the business model it represents remains. Somewhere, another company is building another tool, and the next target could be any country. The question is not whether AI will be used to influence elections. It is whether we will be ready when it is.