Anthropic has uncovered an AI-powered influence operation that allegedly uses its Claude chatbot to target Malaysian voters, according to the company’s latest threat intelligence report. The September 2026 report examines how AI is being used for malicious activities, including cyberattacks, influence operations, scams, surveillance, biological research and attempts to replicate AI models.
The company identifies the case as “GTG-84005”, describing it as a commercial influence operation primarily targeting users in Malaysia. Anthropic says the operation is linked to BBS Bilisim Teknolojileri, a technology company based in Istanbul, Turkiye, which allegedly marketed an “influence-as-a-service” platform.
AI Platform Profiles Malaysian Voters
According to Anthropic, the platform uses Claude to build a constituency-level targeting system based on census and electoral data, covering all 222 parliamentary constituencies in Malaysia. It was reportedly designed to identify political and social faultlines involving sensitive issues such as race, religion and royalty.
The operation also maintained roughly 1,000 fake X accounts, with mechanisms designed to make them appear legitimate and evade platform detection. Anthropic says the platform could also be used to artificially inflate engagement, including a request to generate one million artificial views for the account of Malaysia’s sitting prime minister, although it could not independently verify the figures reported by the operators’ own tools.
Beyond social media manipulation, the operation allegedly ran a synthetic news outlet called “Malaysia Pulse”. Anthropic says Claude was used to scrape legitimate Malaysian news reports, rewrite them and republish the material under fabricated bylines, while content from Russian and Chinese state-aligned outlets was also repackaged as apparently independent Malaysian reporting.
The company also discovered fabricated intelligence dossiers containing false allegations against a Malaysian opposition politician and civil-society organisations. Anthropic says Claude refused some requests linked to the activity, but the operators attempted to negotiate alternative wording with the model to continue pursuing the same objectives.
Anthropic rates GTG-84005 as Category Two under the Brookings Institution’s Breakout Scale, meaning its assets were distributed across multiple platforms but the company found no evidence that the operation’s content had broken out into authentic communities.
Interestingly, Anthropic says the operators also pursued a contract with Malaysia’s national communications regulator. However, it found no evidence that the attempt was successful.
One Of Several AI Misuse Cases
This is among one of several cases highlighted in Anthropic’s report. Other findings include Russian-linked cyber espionage campaigns using Claude, AI-assisted influence operations producing fabricated articles, state-aligned propaganda activity, attempts involving biological research and weapons development, and efforts by Chinese AI companies to use Claude interactions to reproduce its capabilities.
Anthropic says it disrupted the operations it identified, and removed accounts involved in the activity. It added that the information gathered from its investigations is used to strengthen its safeguards and share relevant intelligence with industry partners and authorities.
Transparency Comes With Its Own Questions
Ironically, while Anthropic is using reports such as this to demonstrate greater transparency around how its AI can be misused and to help prevent similar activity, the company itself is facing scrutiny from users over how it collects and uses information. That has raised a separate question over whether efforts could come at the expense of user privacy, particularly as companies gather more data to understand how their models are being used.
These issues are not necessarily the same, but they highlight the growing scrutiny faced by AI companies. While they are expected to be transparent about AI misuse, they are also facing questions over how they collect and use user data.
(Source: Anthropic [official website])