New Delhi: A fake news network operating run out a district in northern Bangladesh used an Anthropic AI chatbot to mass-produce false Bengali-language stories, and part of what it published aligned with pro-Indian interests, AI company Anthropic said in a report released Thursday.
The network promoted Bangladesh’s Awami League and attacked the party’s rivals, according to the report, which the company tagged internally as GTG-54006. “Some content narratives created by the network also aligned with pro-Indian geopolitical interests,” it said, adding that investigators found no proof of state direction or funding behind the activity. Across the report’s 154 pages, it is the only line that refers directly to India.
The pro-India overlap sits against the recent history of the two countries’ ties. The Awami League was India’s principal partner in Bangladesh during Sheikh Hasina’s years in office. Hasina left the country during the student-led uprising of July and August 2024 and has remained in India since.
A caretaker administration governed until 12 February 2026, when Bangladesh held its first general election after her removal; the Awami League was kept off the ballot after its registration was suspended, and the Bangladesh Nationalist Party, led by Tarique Rahman, formed the government. Dhaka has since asked India to extradite Hasina, who was sentenced to death in absentia in November 2025 over the 2024 crackdown, and India has said the request is under examination.
“Because the Awami League was out of power throughout the period the network operated, the activity was carried out on behalf of an opposition party rather than a government,” the report said.
The report, ‘Detecting and countering misuse of AI: September 2026’, documents cases in which people used Anthropic’s Claude models for harm between December 2025 and August 2026.
One operator, 29 accounts
Anthropic attributed the Bangladesh operation to a single actor in Gaibandha district, who used 29 Claude accounts over about 16 months to stay within platform limits and avoid detection.
The person used a programme named “fake_news_3.py” to produce content in fixed runs of 15 headlines, three fabricated stories and 15 image prompts. By the company’s count, the operation produced at least 1,500 headlines, 300 false narratives and 1,500 image prompts. Claude does not generate images, so those prompts were most likely sent to other AI systems to create the visuals.
The output fed a cycle of Facebook Live, YouTube and TikTok streams aimed at Awami League supporters in rural areas with limited literacy. A second script uploaded videos to YouTube on a schedule set a month in advance, routed through a third-party service to mask the operator’s location.
The report said the operator knew the content was false. In his own notes, cited by Anthropic, he wrote that “no one knows the news is fake”. The company said the output was calibrated to be “hot and aggressive” and written so that “even village people understand”, and that the audience was targeted on the assumption it believed the content was real news.
The fake narratives
Anthropic set out the themes the network pushed.
Awami League opponents, including the BNP, Jamaat-e-Islami and the National Citizens Committee, were portrayed as planning Taliban-style rule and infiltrating the security forces.
Other stories fabricated assassination plots against the caretaker government and student-protest leaders, described those leaders as foreign-intelligence agents, and alleged corruption and secret cross-party alliances.
The videos appeared on several Bangladesh-focused channels across all three platforms, the report said. Anthropic said it could not identify every account that published them and found no evidence the content reached audiences beyond that circle. On its Breakout Scale, it rated the operation Category Three.
The company said it identified the activity through internal investigations, banned the accounts, built detection around the network’s behavioural signatures, and shared indicators with the platforms.
The wider report
The Bangladesh case is one of several in a report covering seven areas of misuse: Cyber operations, influence campaigns, surveillance, weapons development, biological misuse, scams, and the theft of AI models.
Anthropic described the fake news cases as its most notable to date, involving suspected state-sponsored groups, criminals, spyware vendors, propaganda bodies and individuals. Other cases include fake dating apps built to defraud users and surveillance tools designed to track dissidents.
The report said that “sophisticated attacks no longer require sophisticated attackers”, and that AI had narrowed the gap between well-resourced state operations and individual operators, allowing one person to run campaigns that would previously have required a team.
The misuse involved Claude’s Haiku, Sonnet and Opus models. The company said none of it involved its Fable or Mythos systems, apart from one case of model theft.
Anthropic said it published the cases because it considers it a responsibility to disclose misuse of its services, and said the risks would grow as models became more capable “unless AI developers and society’s defenders act to make them safer”.
The report does not name any Indian person, organisation or platform, and does not allege Indian involvement in the network. Its India reference is limited to the statement that part of the network’s output aligned with pro-Indian interests, which the company said it recorded without evidence of who directed it.
(Edited by Viny Mishra)
Also read: ‘This isn’t a marketing stunt. AI could kill all humans.’ Anthropic researcher quits, sounds alarm