Fabricated news production targeting Bangladeshi audiences: GTG-54006 case

What happened

We identified and removed an ongoing automated disinformation network that used Claude to generate fabricated Bengali-language news in Bangladesh. The operation focused on promoting the Bangladesh Awami League party and attacking its opponents. Since the Awami League has not been in power since the July 2024 uprising, the network's activities were carried out on behalf of the opposition rather than the government. The actor was fully aware of its deceptive methods, writing in internal communications that "no one knows these news are fake."

A single actor based in Gaibandha District, Bangladesh, ran the operation, rotating 29 Claude accounts to evade platform limits and detection. The actor used a custom software program called 'fake_news_3.py' to connect directly to Claude, which was coded to generate content in fixed batches: 15 headlines, 3 detailed fabricated reports, and 15 image generation prompts. According to the actor, the output was intended to feed a continuous loop of Facebook Live, YouTube, and TikTok livestreams targeting rural Awami League supporters with limited literacy.

The actor generated at least 1,500 headlines, 300 fabricated narratives, and 1,500 image prompts. Since Claude did not yet have image generation capability, these prompts were likely exported to other frontier AI models to create the visual content used in the fabricated narratives.

The operation operated outside Bangladesh, targeting domestic audiences with content designed to serve Awami League interests. Despite this narrative alignment, our investigation found no evidence that the party itself was involved in directing or funding the network's activities.

What the AI did

Claude was used to mass-produce headlines, fake news, and image prompts according to templates. The report notes that Claude did not have image generation capability at the time, so these prompts were most likely taken to other models to produce images.

This public account does not reproduce any fake headline or story, so as not to feed the false narrative back into search results.

What the report observed

Our investigation showed that videos from the operation appeared across multiple Bangladesh-themed channels and accounts on all three social media platforms. We could not identify the specific channels that published all of the output. In addition, we found no evidence that the content reached a broader audience beyond these accounts.

Using the Breakout Scale, we assessed the operation as Tier 3 (multiple platforms, with videos matching the operation's output observed across multiple Bangladesh-themed channels and accounts on social media platforms).

Confirmed & unknown

Platform response

Anthropic removed the relevant accounts.

Limits of response：Removing model accounts cannot delete videos already uploaded to social platforms.