Anthropic Says AI-Powered Operation Targeted Malaysian Voters With About 1,000 Fake Accounts

AI company Anthropic says it has disrupted a commercial influence operation that used its Claude models to help target Malaysian voters with around 1,000 fake social media accounts.

According to a threat intelligence report released on 10 September, the operation used real census and electoral information, including millions of voter records, to build profiles covering all 222 parliamentary constituencies in Malaysia.

However, Anthropic stressed that it found no evidence that the campaign successfully spread into authentic online communities. It also did not identify who paid for the Malaysia-focused operation or link it to a specific Malaysian election.

ADVERTISEMENT

Fake Accounts Targeted Race, Religion and Royalty Issues

Anthropic said the platform used Claude to build constituency-level profiles and focus on politically sensitive issues involving race, religion and royalty.

The network managed more than 1,000 fake accounts on X, formerly known as Twitter.

According to Anthropic, the accounts were designed to appear genuine before being used for influence activities. The system also included measures intended to avoid detection by social media platforms.

A dashboard allowed operators to set targets for artificial engagement.

ADVERTISEMENT

Anthropic said it observed one request intended to generate one million artificial views in support of the account of Malaysia’s sitting prime minister, Anwar Ibrahim.

Importantly, the report did not identify who made that request and found no link between the operation and Anwar or his office. Anthropic also warned that engagement figures recorded in the operators’ own systems could not be independently verified.

ADVERTISEMENT

Operation Also Ran Fake News Site

The operation allegedly went beyond social media accounts.

Read Also:  Student, 19, Investigated After Baby’s Body Allegedly Hidden in Dorm Locker in Malaysia

Anthropic said it ran a fabricated news website called Malaysia Pulse, which took reports from legitimate Malaysian media and used AI to rewrite them before publishing the resulting articles under invented journalist names.

The site also republished material originating from outlets including China’s Xinhua and CGTN and Russia’s Sputnik/RIA and TV BRICS.

According to Anthropic, references showing those organisations’ affiliations were removed so the material appeared to be independent Malaysian reporting.

The company said the Malaysia Pulse domain was registered on 10 May 2026.

Fabricated Dossiers Targeted Politician and Civil Society Groups

Anthropic also said Claude was used while creating purported intelligence dossiers containing false allegations against an opposition politician and civil society organisations.

The company did not name the people or groups targeted.

At several points, Claude refused or partly refused requests from the operator. In one case, Anthropic said the model recognised that a document appeared intended for political defamation.

The operator then altered the wording of its requests while continuing to develop the broader system, according to the report.

Anthropic said it traced the operation to links with BBS Bilisim Teknolojileri, an Istanbul-based technology company that it said sold access to the platform as a commercial influence service.

Anthropic identified and removed the Claude account involved and said it used information from its investigation to map more of the network.

Malaysian Regulator Says It Was Not Involved

Anthropic’s report also said the operator had tried to pursue a contract with Malaysia’s national communications regulator, although Anthropic found no evidence that the effort succeeded.

Read Also:  Cinnabon to Close Last Singapore Outlet at ION Orchard on 22 Sep

Later on 11 September, the Malaysian Communications and Multimedia Commission, or MCMC, said it was taking the report seriously and examining the issues raised.

MCMC said the activities described by Anthropic were not part of its functions or mandate and that it had not been involved in the activities or arrangements outlined in the report.

The regulator said it would seek more information and clarification from relevant parties before deciding whether further action within its jurisdiction was necessary.

Anthropic classified the campaign as Category Two on its “Breakout Scale”. In practical terms, the company said the operation had established assets across several platforms, but it had not found evidence that those assets gained traction within genuine communities.