Fake News Created Using AI in Support of Awami League for 16 Months

A US-based artificial intelligence company, Anthropic, has identified and shut down an automated network that used AI to create fake news in Bengali. According to the company’s investigation, a person from Gaibandha district in Bangladesh had been creating fake news in support of the now-banned Awami League for 16 months. During this period, he used 29 Claude accounts to generate prompts for at least 1,500 fake headlines, 1,500 fake images and 300 fabricated reports.

Anthropic disclosed the information in a report titled “Detecting and Countering Misuse of AI: September 2026,” published on its website last Thursday. The report details several cases of AI misuse in different countries. The Bangladesh-related case was identified by the company as “GTT-54006.”

Anthropic said the content produced by the network targeted the political rivals of the Awami League. However, the company found no evidence of any organisational connection behind the operation or its funding. According to the Anthropic report, the person from Gaibandha repeatedly used 29 Claude accounts to avoid detection and platform restrictions. He also developed his own software. The software was directly connected to Claude and generated content according to a specific format.

The software generated 15 fake headlines, three detailed fabricated news reports and 15 prompts for creating images at a time. According to Anthropic, at least 1,500 headlines, 300 fabricated narratives and 1,500 image-generation prompts were produced in this way. Since Claude did not have its own image-generation feature, Anthropic believes the prompts may have been used with another AI model to create the images.

The report said the person internally described the content he produced as “fake news.” He wanted the content to be “provocative” and “aggressive” while using simple language so that even people in rural areas could understand it. According to the company, the content was designed to make viewers believe that it was genuine news. The content primarily targeted Awami League supporters in rural areas with limited literacy.

The content was created for use in Facebook Live, YouTube and TikTok livestreams. After being generated, the content was stored in specific cloud-storage folders. It was later converted into audio and video. The person also developed a separate script using YouTube’s API to upload videos in batches. Anthropic said a third-party continuous integration service allowed the videos to be scheduled for publication in advance, even up to one month earlier.

One section of the report said the third-party service used for the uploading system also provided a way to hide the person’s IP address.
Anthropic said the content produced by the network was “uniformly” pro-Awami League. The content targeted the BNP, Jamaat-e-Islami, the National Citizens’ Committee (now the NCP), the interim government and leaders of the student movement.

The report identified four major narratives or tactics used in the fake campaign.
First, under a narrative of religious extremism, the BNP, Jamaat and NCP were portrayed as planning to establish Taliban-style Sharia-based rule and infiltrate the security forces.
Second, under narratives of violence and conspiracy, fabricated stories about alleged assassination plots and “hit squads” targeting leaders of the interim government and the student movement were spread.
Third, student leaders of the July movement were portrayed as agents of foreign intelligence agencies.
Fourth, opposition parties were accused of fabricated financial corruption and secret alliances among political parties.

Anthropic said some of the narratives produced by the network were also consistent with India’s geopolitical interests. However, the investigation found no evidence that the activities were directed
or funded by any state. According to Anthropic’s “breakout scale,” the incident was classified as Category Three. This means videos matching content produced by the network were found across various channels and accounts in Bangladesh on multiple social media platforms.

The company said that after identifying the network through an internal investigation, it banned the Claude accounts involved. Anthropic has also developed a system to identify the behavioural patterns associated with the person’s activities because of concerns that he may create new accounts and attempt to carry out similar operations in the future.