A Malaysia-focused AI-driven voter profiling system was built using census and electoral data to map all 222 parliamentary constituencies, aiming to exploit fault lines around race, religion, and royalty.

Anthropic disclosed that a Malaysia-targeted election-manipulation operation used Claude AI models to profile and target voters by constituency across the full set of districts, aided by about a thousand fake X accounts, a fake news outlet, and fabricated dossiers.

The operation operated between late 2025 and mid-2026, disrupting a commercial manipulation platform that spread across multiple platforms and sought to influence voters nationwide.

Reports indicate the operator pursued a contract with Malaysia’s national communications regulator, though there is no evidence the deal went through and details were not disclosed.

Fabricated intelligence reports were created to smear an opposition politician and civil society groups, later altered to bypass defamation checks.

Anthropic found the campaign did not breach authentic online communities; assets were distributed across platforms but failed to penetrate real communities.

All accounts tied to the group have been terminated by Anthropic, which updated safety filters and shared threat intelligence with authorities and partners.

A fabricated media ecosystem, including a fake outlet and an inauthentic YouTube channel named Malaysia Pulse, republished local articles and content from state-aligned foreign outlets while omitting state attribution.

The bogus Malaysia Pulse site scraped legitimate reporting and rebranded content from Xinhua, CGTN, Sputnik/RIA, and TV BRICS as independent reporting.

The operation used automated bots with evasion tactics (renews cookies and IPs) and managed engagement manipulation, including a claimed one-million-view request for the prime minister’s account, with no link found to the PM or his office.

Over 1,000 fake X accounts with warm-up mechanisms and a dashboard to regulate artificial views were used, aimed at influencing targets while evading detection.

The report appears in Anthropic’s September threat intelligence briefing on malicious uses of Claude AI, alongside other alleged Russia-linked campaigns and AI-exploitation cases.