Play store
A Bangladesh-based influence operation used artificial intelligence to produce at least 300 fabricated news reports and 1,500 image-generation prompts in Bangla over about 16 months, according to AI company Anthropic.
The operation, which Anthropic identified as “GTG-54006”, was allegedly run by a single individual in Gaibandha who rotated among 29 Claude accounts to generate political content supporting the Awami League and targeting its opponents.
Anthropic disclosed the findings in its report, Detecting and Countering Misuse of AI: September 2026, published on Thursday.
The company said the operation primarily targeted Bangladesh’s rural population. As the Awami League has been out of power since the July 2024 mass uprising, Anthropic assessed that the campaign was not aimed at supporting a government but was conducted from an opposition political position.
29 accounts used to generate content
According to Anthropic, the operator used custom software called “Fake_News_3.py”, which was directly connected to Claude. The software generated batches containing 15 headlines, three detailed fabricated reports and 15 image-generation prompts.
Anthropic said it found evidence that at least the third version of the software had been used.
The company estimated that the operation generated at least 1,500 headlines, 300 fabricated reports and 1,500 image-generation prompts. As Claude does not generate images, Anthropic said the prompts may have been used with other AI models.
The material was initially stored in a designated cloud folder before being converted into audio and video content for systematic distribution on Facebook, YouTube and TikTok.
Four main disinformation narratives
Anthropic identified four major narratives in the operation.
One portrayed the BNP, Jamaat-e-Islami and National Citizen Party as planning to establish Taliban-style rule and infiltrate security forces.
Another involved fabricated reports about assassination plots and hit squads targeting leaders of the then-interim government and members of the student movement.
A third portrayed student leaders of the July uprising as agents of foreign intelligence agencies.
The fourth involved fabricated allegations of corruption and secret political alliances against opposition parties.
Anthropic said internal communications indicated that those involved were aware they were producing false information. The content was designed to be aggressive, easy to understand and credible to people with limited access to education.
Automated distribution
The operation also used technology capable of bulk-uploading videos through YouTube’s API.
The upload script was operated through a third-party continuous integration service, allowing videos to be scheduled up to a month in advance. Anthropic said the service could also conceal the actual operator’s IP address.
Some of the narratives were consistent with pro-India geopolitical interests, but Anthropic said it found no evidence of state direction or funding.
The operation was placed in the third category of Anthropic’s “breakout scale” classification after similar videos were found on multiple Bangladesh-focused Facebook, YouTube and TikTok channels and profiles.
However, the company said it could not establish which channels had published the complete range of material or find evidence that the content had spread widely beyond the identified accounts.
Claude accounts shut down
Anthropic said it shut down all Claude accounts linked to the operation after identifying the network through an internal investigation.
It also developed a detection system based on the operation’s behavioural patterns to identify and prevent similar attempts. Relevant evidence was shared with social media platforms and other partners, although some technical details were withheld from public disclosure for security reasons.
Anthropic’s September report details nine cases involving the misuse of AI in influence operations, cyberattacks, propaganda and other activities across several regions.
The Bangladesh case was unusual, the company said, because a single individual, rather than a state institution, used AI to conduct a relatively large-scale political content-generation operation.
Anthropic said AI is increasingly allowing less-skilled individuals and small groups to conduct disinformation campaigns and cyberattacks at greater scale. It called for stronger information sharing and coordinated security measures among governments, researchers and AI companies.