"You can flood the network": How AI threatens to change elections

A report by Anthropic reveals a political influence operation in Malaysia, where the Claude AI model was used to create a network of about 1,000 fake accounts, a fictitious news site, and fabricated documents, in an attempt to influence voters in the recent elections. Experts warn that AI enables the spread of disinformation on an unprecedented scale.

A report by the artificial intelligence company Anthropic, published in September, reveals a widespread political influence operation in Malaysia, in which the Claude model was used to create an entire infrastructure for spreading disinformation. The operation included a network of about 1,000 fake accounts on X, a fictitious news site named "Malaysia Pulse," and a series of fabricated documents. According to the company, the operation aimed to influence voters in the recent Malaysian elections, exploiting sensitive rifts in Malaysian society around issues of race, religion, and monarchy, across all 222 parliamentary constituencies. The operation failed after Anthropic's intervention, but its exposure raised serious concerns among experts about the use of AI in future election campaigns in Southeast Asia. The fake site copied real reports from Malaysian media outlets and rewrote them under fictitious bylines, and also republished articles from foreign media outlets associated with Russia and China while removing attribution to the source. Experts note that AI makes the process of spreading disinformation much simpler and faster, and allows for the creation of personalized content on an unprecedented scale.