Anthropic has disclosed that its large language model Claude was leveraged in a coordinated political influence scheme in Malaysia. The operation assembled roughly one thousand accounts on the X platform, launched a fabricated news site called Malaysia Pulse, and produced a series of counterfeit documents. This marks a shift from labor-intensive disinformation campaigns to ones powered by generative AI tools.
The campaign focused on recent state elections, drawing on census and voter data to exploit sensitive divisions such as ethnicity, religion and royal affiliations across all 222 parliamentary constituencies. Anthropic intervened after the activity was detected, rendering the effort ineffective. The episode has heightened concerns that similar AI-driven tactics could be deployed in upcoming national contests throughout the region.
In a threat assessment released in September, Anthropic reported that the actors behind the Malaysian operation included governments, state-aligned propaganda bodies and private firms selling influence. They built networks of counterfeit social-media profiles and news sites that reached audiences on six continents, often republishing material from Russian and Chinese state outlets while stripping original bylines. The company says it disrupted the activity, improved its safeguards and shared intelligence with relevant authorities.
Former Meta public-policy official Katie Harbath described the case as evidence that AI adds a new level of sophistication and ease to the disinformation ecosystem. In Malaysia, media lecturer Dr Ben Loh noted that AI-generated content reinforcing hostility toward Rohingya refugees simply amplifies pre-existing prejudice. Both observations underline the potential for generative models to magnify societal fault lines when weaponised.
The region’s vulnerability to fake news has been highlighted by figures such as Nobel laureate Maria Ressa, who called the Philippines a “petri dish” for disinformation experiments later copied elsewhere. OpenAI responded to abuse in the Philippines by banning accounts that generated political commentary supporting President Ferdinand Marcos Jr. Meanwhile, AI has also been employed positively, as an AI-created cartoon of Indonesia’s Prabowo Subianto helped soften his strong-man image during the 2024 election.
Upcoming polls in the area include a possible Malaysian general election in late 2027, national votes in the Philippines and Cambodia in 2028, and Indonesia’s election the following year. Harbath warns that future campaigns could attempt to “flood the zone,” overwhelming large language models with tailored propaganda to shape the information they later provide. Such tactics could further erode electoral integrity across Southeast Asia.