AI, Elections and Corporate Power: Reading the Anthropic Report from Indonesia

From Project Multatuli, Original writer Antonia Timmerman, October 16, 2026

Latest Anthropic Report

Anthropic’s latest report shows how AI is being used for cybercrime, surveillance, and political manipulation. The findings are important for Indonesia ahead of the 2029 elections, but they also invite critical questions: to what extent should AI companies determine for themselves what constitutes abuse and threats?

The company that owns the AI service, Anthropic, released a report on the various abuses of Claude it detected in the period December 2025 to August 2026. These reports are useful on several levels from the most obvious to the not-so-obvious.

Manipulate Public Opinion

First, the most obvious one: it gives us an idea of how bad actors use LLMs to stalk, deceive, hack, and manipulate public opinion. In addition to financially motivated actors, we can also see how LLM is integrated into the set of tactics orchestrated by state actors to monitor and control their citizens.

In a number of separate operations, actors affiliated with the Chinese government used Claude to monitor various elements of society, including religious leaders, Hong Kong pro-democracy activists, and Uyghur advocacy groups.

Criminals

Criminals also use Claude for extortion and data theft. Anthropic noted several operators connected to a collective called ShinyHunters. They commit massive data theft, threaten the victim and then ask for a ransom so that the data is not spread on the dark web.

Anthropic also identified several operators in China — two of whom were students — who were spying on private organizations and government agencies in the Middle East, Europe, and Southeast Asia.

AI has cut the cost of cyber operations significantly. Major operations can now be performed by one or a handful of people.

Greater Speed Scale and Depth

However, keep in mind that almost all of the attack operations identified by Anthropic were conducted before the LLM was publicly accessible. What makes the difference is how malicious tactics are now being implemented at a much greater level of speed, scale, and depth.

According to Center for Strategic & International Studies (CSIS) researcher Nikita Shah, there are actually very few types of threats that are completely new in the Anthropic report. Moreover, the existential threat of human destruction from AI models that get out of control.

Hacking

In the hacking category, for example, “None of the operations in this report relied on completely new techniques,” Anthropic itself wrote in its report.

If in the past hacking attacks could look sophisticated “due to the large amount of manpower involved now many aspects of the operation have been transferred to AI models … which operate in an integrated way with the speed of the machine and in parallel.”

Indonesia

Thus, existing legal tools can actually be used in dealing with similar cases. In Indonesia, hacking conduct are subject to criminal sanctions under the Information and Electronic Transactions Law (ITE Law) and the Personal Data Protection Law (PDP Law). This is true too for fraudulent conduct.

There are a number of cases from the Anthropic report that are quite worrying for Indonesia, especially ahead of the 2029 presidential election: the manipulation of information and perceptions.

In Malaysia, Claude has been used to create profiles and target voters. Using census data and election data, perpetrators can target voters very precisely, even down to the regional level.

The perpetrators have a network of about a thousand fake X (Twitter) accounts, which are optimized to boost interaction metrics while circumventing social media platform detection systems. In addition, the perpetrators also run a fake news site called “Malaysia Pulse”, and supply content using AI-based writing. They then made false accusations about opposition politicians and civil society organizations.

In these operations, Claude has been used at various stages: managing, running and tracking the network of fake accounts they own as well as recording the interaction metrics they have managed to achieve.

Already BAU

In Indonesia, according to a report by researchers Pradipa Rasidi and Michelle Anindya, most of our digital information landscape is already highly polluted. Engineered by social media algorithms and cyber forces or buzzers that use AI tools. Worse, this engineering activity is a business-as-usual activity, meaning that it has indeed taken place on a daily basis outside of important events such as elections.

So, it would not be an exaggeration to be increasingly careful ahead of the 2029 election, facing an information landscape that is very likely to become even murkier. Read the whole piece here AI, Elections, and Corporate Power: Reading the Anthropic Report from Indonesia – Project Multatuli.