How an AI plot targeted Malaysia’s elections and exposed deep data risks

Anthropic’s discovery of an AI-powered election manipulation platform targeting Malaysia offers a glimpse into how the technology can exploit a country’s political vulnerabilities for sophisticated influence operations.

The operation’s use of millions of voter records, fake social media accounts, a synthetic news site and fabricated dossiers pointed to a deeper risk, analysts said: weak data safeguards and entrenched ethnic and religious fault lines could allow political actors to target voters faster than regulators or platforms can respond.

In a report released on Thursday, the US artificial intelligence firm said it had removed an account that used its Claude model to run what it described as a commercial election manipulation platform across Malaysia’s 222 parliamentary constituencies.

Anthropic said the operators ingested census and electoral data to profile voters constituency by constituency, while using the platform to manage a network of fake accounts designed to inflate online support and avoid detection.