Anthropic, one of the world's leading artificial intelligence laboratories and the developer of the chat boy called Claude, announced in its detailed threat intelligence report titled "Detecting and countering misuse of AI: September 2026" that an Istanbul-based technology company has established a wide artificial intelligence-supported operation network to manipulate Malaysian politics and election processes.
The report stated that the activities recorded under the code "GTG-84005" were marketed as a commercial influence operation platform.
According to the research of the Anthropic Threat Intelligence unit, the Istanbul-based company BBS Information Technologies is at the center of the operation.
While the company portrayed to the public an organization that provides defensive cyber intelligence and counter-disinformation services, Claude misused the artificial intelligence model in the background and sold political influence operation infrastructure to its customers.
In their own documents, the actors described this system as "a military-level, artificial intelligence-driven and real-time political operations ecosystem."
The technical backbone of the operation included fake account networks specially designed to bypass the detection algorithms of social media platforms.
The company has developed control panels through Claude Code where these fake profiles are managed, tracked and impact analyzed.
The system managed more than a thousand fake X (Twitter) accounts equipped with "warming algorithms" that periodically refreshed cookies and IP addresses, allowing them to behave like real users for a certain period of time.
Artificial interaction was directed to target accounts through these developed panels. It was seen that the system records in the report included a request to transfer 1 million artificial views to the leader's social media account in order to support the incumbent Malaysian Prime Minister Anwar Ibrahim.
Their dashboard was found to have recorded millions of fake views and likes on the account of a senior government official; However, Anthropic noted that the figures could not be confirmed by independent sources because the data in question were the actor's own panel outputs.
In order to manipulate public opinion, the disinformation network launched an artificial news site called "Malaysia Pulse" and its associated YouTube channel, whose domain name was registered on May 10, 2026.
The system automatically scanned the news of legitimate Malaysian media outlets and reprinted them multiple times through Claude, publishing them with fictitious author names.
The fake media in question also functioned as a means of circulating the content of official media organs affiliated with the Chinese and Russian states.
News from channels such as TV BRICS, Xinhua, Sputnik and CGTN were presented as original content of the independent Malaysian press, with official state affiliations removed from the text.
It was reported that the company also tried to sign a commercial contract with the Malaysian national communications regulatory authority in order to increase the legitimacy of the system, but there was no evidence that this initiative achieved its goal.
Anthropic revealed the technical infrastructure of the operation in detail. During the investigations, the network's server IPs hosted on Hetzner, various domain names and the software repository on GitHub were directly reported.
A large number of fake accounts connected to the system, opened all at once on May 17, 2026, were also recorded in the detection records.
The company announced that it had permanently closed all accounts connected to the network, disabling the operation in its early stages.
Researchers stated that the content produced could not be disseminated to independent audiences and classified the impact of the operation as Category 2 according to the Brookings Institution's Breakout Scale.
This degree shows that the content is distributed in different media but does not find an organic response in real segments of society.
Anthropic was founded in San Francisco, USA, in 2021 by sisters Dario Amodei and Daniela Amodei, former research directors of OpenAI.
The founders of the company left OpenAI on the grounds that commercial concerns pushed artificial intelligence security and ethical principles into the background.
The company, which determines its main purpose as developing controllable, honest and secure systems, receives billions of dollars of investment support from technology giants such as Google and Amazon.
The multi-modal artificial intelligence assistant called Claude, developed by the company, performs visual, audio and complex data analysis in addition to text production.
Instead of human intervention, the model controls its own responses using principles called "Constitutional Artificial Intelligence" and can process hundreds of pages of text at once thanks to its wide context window.
Offering three main versions for different needs: Haiku, Sonnet and Opus, the system provides services in a wide range of areas, from code writing to strategic analysis, directly or through corporate cloud integrations.
