One person ran a pro-Awami League ‘fake news’ network using AI: Anthropic
A US artificial intelligence safety company has uncovered an automated disinformation operation run by a single operator from Bangladesh’s Gaibandha district, which produced hundreds of fabricated political stories and thousands of false headlines for distribution on social media
Anthropic said the operation, identified as GTG-54006, was active for around 16 months and involved 29 accounts producing Bengali-language content targeting audiences in Bangladesh.
The operator used artificial intelligence tools and automated software to generate political content, including false allegations, fabricated assassination plots and claims targeting opposition politicians and figures associated with the interim government.
According to Anthropic, the operator used a Python programme called "fake_news_3.py" to generate batches of content. Each run produced 15 Bengali headlines, three fabricated news stories and 15 prompts for generating images.
The company said the operation generated at least 1,500 fake headlines, 300 fabricated news stories and 1,500 image-generation prompts.
The content was then converted into videos and distributed through platforms including Facebook, YouTube and TikTok. The operator also used cloud services to schedule YouTube uploads in advance, according to Anthropic.
Political narratives
Anthropic said the operation became active in the aftermath of Bangladesh’s July 2024 uprising and generated content that generally promoted narratives favourable to the Awami League while attacking political opponents.
Among the narratives identified by the company were false claims that BNP, Jamaat-e-Islami and the NCP were infiltrating security forces to establish Taliban-style Sharia rule.
The operation also produced fabricated reports about alleged assassination plots targeting leaders of the July movement and officials of the interim government. Other false claims portrayed figures associated with the protests as foreign intelligence agents and alleged secret political deals and financial corruption involving political groups.
Some of the narratives also appeared to align with pro-India geopolitical positions, Anthropic said.
However, the company found no evidence that the Awami League leadership funded, directed or was aware of the operation. It also found no evidence of funding or direction from any foreign government.
Accounts shut down
Anthropic said it identified the operation through its internal monitoring and subsequently banned the accounts associated with it.
The company said the case demonstrates how AI tools and automation can enable a single individual to produce and distribute political disinformation on a scale that would previously have required a much larger operation.
The findings also highlight the growing challenge for social media platforms and users in identifying AI-assisted disinformation, particularly as political content increasingly spreads through platforms such as Facebook, YouTube and TikTok.