TDS Desk:

A single operator in northern Bangladesh used Anthropic’s Claude AI chatbot to mass-produce fabricated Bengali-language news stories that promoted the Awami League, attacked its political rivals and, in some cases, aligned with pro-Indian geopolitical interests, according to a report by the US artificial intelligence company released yesterday (10 September).

The operation, which Anthropic identified internally as GTG-54006, was run from Gaibandha district and used 29 Claude accounts to produce at least 1,500 headlines, 300 fabricated stories and 1,500 image prompts, according to the report, “Detecting and countering misuse of AI: September 2026.”

The company, however, found no evidence that the operation was directed or funded by any government, reports The Print.

The company said the operator used Claude over about 16 months, switching between accounts to stay within platform limits and avoid detection.

The activity was documented as part of Anthropic’s review of harmful uses of its AI systems between December 2025 and August 2026.

How the operation worked

According to Anthropic, the operator used a programme called “fake_news_3.py” to generate content in fixed batches of 15 headlines, three fabricated stories and 15 image prompts.

Because Claude does not generate images, the prompts were likely fed into other AI systems to create visuals, the report said.

The material was then used in Facebook Live, YouTube and TikTok content targeting Awami League supporters, particularly in rural areas with limited literacy.

A second script automatically uploaded videos to YouTube according to a schedule set a month in advance. The operation also used a third-party service to conceal the operator’s location, according to the report.

Anthropic said the operator knew the material was false. Notes recovered during the investigation included the line, “no one knows the news is fake”.

The company said the content was designed to be “hot and aggressive” and written so that “even village people understand”, reflecting an effort to make the fabricated stories appear accessible and credible to the target audience.

What the fake stories claimed

The network portrayed opponents of the Awami League, including the Bangladesh Nationalist Party (BNP), Jamaat-e-Islami and the National Citizens Committee, as planning Taliban-style rule and infiltrating the security forces, according to Anthropic.

Other fabricated narratives alleged assassination plots against members of the caretaker government and leaders of the student movement, portrayed those leaders as agents of foreign intelligence services, and accused political figures of corruption and secret alliances across parties.

The videos appeared on Bangladesh-focused channels across Facebook, YouTube and TikTok, the report said.

Anthropic said it could not identify every account involved in publishing the material and found no evidence that the content reached audiences beyond that circle. It classified the operation as Category Three on its Breakout Scale.

The company said it identified the activity through internal investigations, banned the associated accounts, developed detection methods based on the network’s behavioural patterns and shared relevant indicators with the platforms.

No evidence of Indian direction

Anthropic’s report refers once to content aligning with pro-Indian geopolitical interests.

The company did not name any Indian individual, organisation or platform, and did not allege that India was involved in directing or funding the operation.

The finding comes against the backdrop of close ties between India and the Awami League during Sheikh Hasina’s years in power. Hasina left Bangladesh during the July-August 2024 uprising and has remained in India since.

The report said the operation was conducted on behalf of an opposition party rather than a government because the Awami League was out of power throughout the period examined.

One person, a large-scale influence operation

Anthropic described the Bangladesh case as one of its most notable examples of AI-assisted fake news activity to date.

The broader report covers seven areas of misuse of its models: cyber operations, influence campaigns, surveillance, weapons development, biological misuse, scams and AI model theft.

Other cases documented in the report include fake dating applications designed to defraud users and surveillance tools intended to track dissidents.

Anthropic said the cases showed how AI was narrowing the gap between well-resourced operations and individual actors, allowing one person to conduct campaigns that would previously have required a larger team.

The misuse involved Claude’s Haiku, Sonnet and Opus models. Anthropic said none of the cases involved its Fable or Mythos systems, apart from one incident involving model theft.

The company said it disclosed the cases because it considers documenting misuse of its services a responsibility, warning that the risks would increase as AI systems become more capable unless developers and others work to make them safer.