In a striking revelation that underscores the growing intersection of artificial intelligence and geopolitical manipulation, US-based AI company Anthropic announced it has dismantled a sophisticated, commercial-grade disinformation network that was actively targeting the Malaysian electorate. The operation, which leveraged the company’s own Claude AI models, represents a new and concerning frontier in digital warfare, where advanced technology is weaponized not by rogue states or shadowy hacker collectives, but sold as a service to the highest bidder. The disclosure, detailed in a threat intelligence report on September 10th, peeled back the curtain on a hidden campaign designed to manipulate public opinion and interfere with the democratic process in Malaysia, highlighting the novel vulnerabilities that AI introduces into the already complex landscape of modern information warfare.
At the heart of this digital subterfuge was a network of approximately one thousand fake social media accounts on the platform X, a fabricated news outlet, and a series of manufactured intelligence dossiers. Anthropic’s investigators traced the operation’s infrastructure to an Istanbul-based technology company, BBS Bilisim Teknolojileri, which was allegedly marketing this entire apparatus as a paid “influence-as-a-service” tool. The company’s own promotional materials boasted of a “military grade, AI-driven, real-time political operations ecosystem,” a chillingly clinical description of a system built to deceive. This wasn’t a crude, lone-wolf operation; it was a polished, professional product designed for clients seeking to covertly steer public perception. The scale and audacity of the setup, with its own dashboard and performance metrics, paints a picture of a shadowy industry emerging to exploit the power of generative AI for political ends, with Malaysia turning out to be a key battleground for this new form of covert action.
The operational methodology deployed against Malaysia was particularly insidious, as it was engineered to exploit the nation’s most sensitive social and political fault lines. According to Anthropic’s report, the actor utilized Claude to construct an AI-powered voter profiling platform that ingested comprehensive census and electoral data from all 222 parliamentary constituencies across the country. This allowed the operators to micro-target specific demographics with tailored messaging, deliberately focusing on divisive issues surrounding race, religion, and royalty. To lend an air of legitimacy to their fabricated content, the network controlled a fake media outlet and an inauthentic YouTube channel branded as “Malaysia Pulse.” This front organization would use AI to scrape and rewrite legitimate local news articles under fictional bylines, creating a veneer of independent journalism. Furthermore, it would republish articles from state-aligned foreign outlets like Russia’s Sputnik and China’s Xinhua, stripping them of their original attribution to present them as independent Malaysian reporting, thereby seeding foreign narratives into the local discourse under the guise of homegrown journalism.
This Malaysian operation was not an isolated incident but rather a component of a broader global pattern of AI-enabled influence operations uncovered by Anthropic. The report connects the same tactics to Russian propaganda efforts in Moldova and Africa, the demographic profiling of citizens in Iran and China, and the amplification of AI-generated content across a sprawling network of 70 fake outlets operating in 20 languages through coordinated social media botnets. This global perspective reveals a disturbing trend: the democratization of sophisticated disinformation capabilities. The same foundational technology designed to be helpful and creative is being repurposed by a new class of actors to conduct large-scale, cross-border manipulation with an efficiency that was previously unimaginable. The case studies presented in the report span a wide range of harms, including surveillance, financial scams, and even attempts at biological misuse, painting a comprehensive picture of the challenges facing AI safety researchers as they race to anticipate and counter novel forms of abuse.
The technical sophistication of the operation extended to its evasion tactics. The thousand automated X accounts were equipped with “warm-up” mechanisms, designed to simulate authentic user behavior by regularly renewing cookies and IP addresses to slip past platform detection systems. The platform itself offered a user-friendly dashboard, allowing clients to fine-tune the number of artificial views a target should receive. Anthropic’s report notes observing a request in support of the sitting Malaysian prime minister, which called for one million artificial views on his account, though they found no evidence linking the operation directly to the prime minister or his office. Beyond artificial amplification, the platform was also used to draft fake intelligence files, containing false allegations against a Malaysian opposition politician and local civil society organizations. When Claude refused requests that it identified as political defamation, the report details how the operator would engage in “prompt negotiation,” cleverly sanitizing the wording of their requests to bypass the AI’s built-in safety controls, demonstrating a cat-and-mouse game between malicious actors and AI safeguards.
In the wake of the report’s publication, the Malaysian Communications and Multimedia Commission (MCMC) responded with a formal statement, taking the allegations seriously while firmly denying any involvement. The regulator asserted that “activities of the nature described in the report do not form part of MCMC’s functions or mandate, and MCMC does not engage in such activities or arrangements.” This denial, while expected, creates a complex geopolitical ripple, raising questions about the extent of the threat and the potential for foreign interference in Malaysia’s internal affairs. Anthropic has stated that it has terminated all accounts associated with the group, upgraded its safety filters to detect similar prompt negotiation attempts, and shared its threat intelligence with authorities and industry partners. While this is a victory for AI safety, the underlying infrastructure of the “influence-as-a-service” industry may remain, potentially with other clients. The incident serves as a powerful demonstration of the dual-use nature of AI, a tool that can inform and uplift societies, but which can also be used to corrode trust, manipulate public opinion, and destabilize democracies, marking a clear moment of reckoning for both tech companies and governments trying to chart a safe path forward.