Anthropic uncovers pro-Awami League AI fake news network based in Gaibandha
Representational image generated by AI.
A single individual used 29 cloud accounts on a rotating basis to produce fake news using artificial intelligence (AI). The operation was run from Gaibandha, Bangladesh, with the aim of promoting the Awami League.
US-based AI company Anthropic said it had identified and shut down an automated network involved in producing fake news in Bangla. The company disclosed the information in a report published on September 10 on efforts to prevent the misuse of AI.
According to the report, the network used Anthropic’s chatbot to generate fake news in Bangla targeting Bangladesh. The content was intended to attack the Awami League’s political opponents and shape public opinion in favour of the party.
According to Anthropic, the individual had developed proprietary software called “Fake News 3.” The software connected directly to the cloud and generated content according to a predetermined format. Each time, it produced 15 fake headlines, three detailed fabricated news reports and 15 instructions, or prompts, for generating images.
Anthropic estimates that the network generated at least 1,500 headlines, 300 fabricated narratives and 1,500 image-generation prompts.
29 Cloud Accounts
According to Anthropic’s report, the individual based in Bangladesh used 29 cloud accounts on a rotating basis over a period of around 16 months. The accounts were switched in this manner to avoid usage limits and detection.
Direct human intervention in the operation was limited. Bangla headlines and news reports were automatically generated using the cloud API. Another script was then used to place the content in designated folders in cloud storage, convert it into audio and video, and publish it on YouTube.
The report said a separate upload script was used to automatically publish videos on YouTube. The publication schedule for the videos could be set as much as a month in advance.
Following its investigation, Anthropic banned the accounts involved. The company also said it was taking additional measures out of concern that the same individual or associated parties might attempt to create new accounts and resume the operation in the future.
Targeting People in Rural Areas
Anthropic said the fake content was primarily created to target people in areas and villages with low literacy rates. The plan was to distribute the content in video format through platforms such as Facebook Live, YouTube and TikTok.
According to the report, members of the network themselves referred to the content as “fake news” in their internal communications. They were instructed to create the language and presentation in an “exciting” and “aggressive” manner so that even people in rural areas could easily understand it.
Anthropic’s investigation found videos linked to the operation on Bangladesh-focused channels and profiles across all three platforms—Facebook, YouTube and TikTok.
However, the company was unable to identify all the specific channels that published the content. It also could not determine how many viewers outside the accounts associated with the videos were reached by the content.
Attacks Targeted BNP, Jamaat, NCP and Interim Government
According to the report, the network spread various forms of misinformation targeting the Bangladesh Nationalist Party (BNP), Jamaat-e-Islami, the National Citizen Party (NCP) and the interim government.
The report said some campaigns portrayed opposition figures as foreign agents and accused them of seeking to establish “Taliban-style rule.”
However, Anthropic said it found no evidence that the Awami League, whose activities are currently banned in Bangladesh, as a party operated, directed or financed the network.
The report said some of the content narratives were also consistent with pro-India geopolitical interests. However, no evidence was found of any state direction or funding behind the operation.
Anthropic’s report presented the Bangladesh case not merely as an instance of political campaigning, but as a new example of the impact of AI-powered automated systems on the spread of misinformation.
