The video posted by a state branch of India’s ruling Bharatiya Janata Party (BJP) showed Assam chief minister Himanta Biswa Sarma shooting an image of two men in Muslim skull caps. “Foreigner-free Assam”, read one caption across the video. “Why did you not go to Pakistan?” said another.

One of the men in the photo that Sarma was portrayed as shooting was Gaurav Gogoi, a leader of the Indian National Congress (INC), the BJP’s main competitor in Assam for the state’s upcoming legislative elections next month.

Gogoi has stated that he is Hindu but enjoys visiting different religious sites and observing their norms. He has been photographed wearing traditional Muslim attire during religious occasions such as Eid.

But the image of him in the video shared by BJP Assam, wearing a casual singlet with a skull cap, was not one of those occasions.

Bellingcat has seen several dozen videos posted by the BJP that use generative artificial intelligence (AI) alongside anti-Muslim and anti-Bangladeshi messaging in the border states of Assam and West Bengal in December last year, ahead of legislative elections scheduled in both states for April.

Bellingcat analysed 499 social media posts containing photos and videos shared on Facebook, Instagram and X by the BJP’s official accounts in the two states for this time period, finding 194 posts that appeared to meet the United Nations’ definition of hate speech: discriminating against persons or communities based on inherent characteristics such as religion and national origin. Of these, 31 (about one in six of the hateful posts) contained the obvious use of AI-generated imagery.

Chart: Galen Reich

These appear to be part of a larger pattern of politicians and parties globally using generative AI to amplify hateful or divisive content, particularly ahead of major political events such as elections.

Ahead of the New York City mayoral race last year, Andrew Cuomo’s official X account shared, then deleted, an AI-generated video depicting Mamdani eating rice with his hands and a Black man in a keffiyeh shoplifting. In Italy, several opposition parties complained to a communications watchdog after deputy prime minister Matteo Salvini’s League party published a series of AI-generated images depicting men of colour attacking women or police officers. And in the UK, videos by an AI-generated rapper funded by the far-right Advance UK party, with lyrics targeting Muslims, were viewed millions of times.

A Campaign of Hate

Both Assam and West Bengal share a border with Bangladesh. BJP, the world’s largest political party, is currently in power in Assam, where legislative elections are scheduled on Apr. 9. West Bengal, which goes to the polls on Apr. 23, is governed by the Trinamool Congress (TMC).

Tensions between India and Bangladesh worsened after former Bangladeshi Prime Minister Sheikh Hasina, who enjoys close ties with Delhi, was ousted in 2024 and fled to India.

US-based international affairs expert Mohammed Zeeshan told Bellingcat that the “dehumanising and debasing” terminology used in India to refer to alleged illegal Bangladeshi immigrants, including by senior ministers, has caused resentment towards India in Bangladesh.

“The situation, in fact, was so bad that Hasina herself had subtly warned the Modi government in public statements that Indian domestic rhetoric was endangering Bangladeshi Hindus, who bore the brunt of that resentment,” Zeeshan said.

Zobaida Nasreen, a professor of anthropology at Dhaka University, said that anti-Muslim rhetoric intensified by BJP leaders reinforces the belief in Bangladesh that Muslims and Bengalis are being collectively targeted in India.

“Viral videos containing this message tend to spread quickly across Bangladeshi media and social platforms especially on Facebook, enhancing perceptions of hostility and triggering anti-India sentiment or nationalist backlash,” she added.

In December, the month our dataset was collected, Dipu Das, a Hindu garment worker, was beaten to death at an anti-India protest in Bangladesh over allegations that he had made derogatory remarks about Islam.

And while the administration led by Bangladesh’s newly elected leader Tarique Rahman has sought to reset strained ties, most of the hateful social media posts we saw posted by the BJP in December attacked Bangladeshi Muslims and/or Bengali-origin Muslims in India, showing how tensions between the two countries continue to influence political messaging in India’s border states.

Bellingcat’s analysis included a total of 202 posts by BJP Assam and 297 by BJP’s West Bengal branch on their official accounts. We also looked at posts shared by BJP’s main opponent parties – 194 from INC in Assam and 357 from the TMC in West Bengal – during the same time period in December.

This included all visual social media posts (containing photos or videos) by each party in December, except those that did not appear to contain any overt political messaging, such as those simply commemorating public holidays. We only counted each photo or video once, regardless of how many platforms it was shared across.

Although all of the major parties contesting in the Assam and West Bengal state elections appeared to use AI-generated imagery in some of their posts, there appeared to be a particularly high concentration of hateful messaging in the ones posted by the BJP’s accounts.

In Assam, we identified 28 posts by BJP using apparently AI-generated imagery, of which 24 carried hateful messaging. Of the 194 INC posts we looked at from December, 41 appeared to feature AI-generated imagery, but none of these appeared to carry hateful messaging.

In West Bengal, we found 14 BJP posts that contained clear indicators of AI-generated imagery, seven of which were hateful. We also identified 15 posts by the incumbent TMC that appeared to feature AI imagery, but none of these appeared to meet the definition of hate speech.

Support Bellingcat

Your donations directly contribute to our ability to publish groundbreaking investigations and uncover wrongdoing around the world.

When contacted for comment, BJP Assam spokesperson Rupam Goswami did not directly respond to questions on the party’s general use of AI but said they did not post any AI-generated photos of Gogoi. “BJP does not stoop so low,” he told Bellingcat.

As for the “point blank” shooting video, Goswami initially said the person responsible had been punished and removed from the party. However, when asked about Sarma saying that he would re-post the video with those he was depicted shooting labelled as “Bangladeshis”, Goswami said, “[Bangladeshis] need to be completely suppressed.”

BJP West Bengal did not respond to multiple requests for comment by Bellingcat via phone and email.

It is important to note that as generative AI technology improves, it can be increasingly difficult to detect AI-generated imagery. Our manual count of AI-generated imagery only included posts that had obvious signs of generative AI such as unnaturally smooth textures and multiple people with the same faces. It is therefore possible that there were other images in our dataset where generative AI was used more subtly.

However, Joyojeet Pal, Professor of Information at the University of Michigan, told Bellingcat that the quality of these visuals, or whether they looked real, was not the priority.

“What politicians in India have understood is that the sociocultural drivers of misinformation are most important for elections, so they harp on about things to the extent that they have started to not care about form over substance. It looks bad? It doesn’t matter,” he said.

More important to voters, according to Pal, was whether they already believed in the narrative contained in the videos, which generative AI could help create more quickly: “AI is helping cement polarised opinions by giving you the kind of content you have already decided you want to engage with.”

When asked about INC’s use of AI, party spokesperson Aman Wadud said that it was obvious that some of the videos they posted were made with AI and that there was no intention to mislead.

“AI can be both destructive and creative. We are using it in a creative manner, we are not using it in a destructive manner. We don’t violate people’s dignity, we don’t falsely accuse people,” he said.

TMC did not respond to Bellingcat’s multiple requests for comment via phone and email by publication time.

Portraying Bengali Muslims as ‘Foreigners’

The largest category of hateful messaging Bellingcat observed in the BJP’s posts targeted Bangladeshi or Bengali-origin Muslims, referring to them as “infiltrators” or “foreigners”. We counted 66 such posts by the BJP’s Assam and West Bengal branches from December, of which eight appeared to contain obvious AI-generated imagery.

Bengali-origin Muslims are often stereotyped as “illegal immigrants” in the state, although members of the community have lived in India since the late 1800s.

Last year, the BJP deported thousands of alleged undocumented migrants – reportedly including Indian Muslim citizens – to Bangladesh. Human rights groups have called the deportations unlawful and discriminatory, as well as lacking in due process.

One video referencing this theme shows AI-generated visuals of protests against “illegal infiltration” in Assam, with the caption urging people to “wake up” or the country would “turn into Bangladesh”.

A different one uses real footage from past violence in Assam mixed in with images of Muslim men. A song playing in the background accuses them of taking over “Assamese land” and shows AI images of “Assamese” people, i.e. those not in stereotypical Muslim clothing, crying.

Both videos use religious markers to draw a distinction between “infiltrators” – men in skull caps or lungis associated with Bengal-origin Muslims – and “citizens” in non-Muslim attire.

Clothing is often used by the Hindu far-right as a visual shorthand for identity and a deepening religious divide. In 2019, Prime Minister Narendra Modi said of protests against a controversial citizenship law that those responsible for violence could be “identified by their clothes”.

In the hateful posts seen by Bellingcat, both real and AI-generated images of opposition figures – particularly Gogoi – were shown alongside messaging that suggested that they supported “foreigners” or “infiltrators”.

The Center for the Study of Organized Hate (CSOH) also noted, in a 2025 report on AI-generated imagery and Islomophobia in India, that Hindu far-right politicians and media outlets have invoked and reinforced the trope of Muslims as “infiltrators” for years.

“AI-generated images on these themes reinforce associations between Muslim identity and illegality, reinforcing xenophobic and Islamophobic stereotypes. In doing so, they play a powerful role in justifying exclusionary policies and normalising discrimination against Muslims,” the report said.

‘Save Hindus’

Zenith Khan, a data analyst who worked on the CSOH report, noted that AI-generated propaganda was often tightly knit with current political moments, and its impact depended on “timing it right” especially when “people are emotionally charged”.

The violence against the minority Hindu community in Bangladesh has been used by the BJP to raise concerns over the safety of Hindus in India.

Days after Das’ lynching, the Assam state branch of BJP posted a video with an image of his face – except that it was manipulated with AI to show tears streaming from his eyes. “Save Hindus”, said the text accompanying the video.

Posts by BJP’s West Bengal unit also seemed to frame Muslims as criminals or threats. A video, styled after the TV show “Stranger Things”, raised alarms over an “upside down” version of the state under the current government.

A man is depicted being chased by men in skull caps. Arrows label them as “Ralib,” “Galib,” and “Chalib” – a play on Muslim names ending in “-lib” – in case the skull caps left any ambiguity about their Muslim portrayal.

INC filed a police complaint in September last year against the BJP for sharing AI videos targeting Gogoi and the Muslim community, as well as another complaint in relation to the video of Sarma portrayed as shooting two men “point blank” in February.

INC Assam spokesperson Wadud said that no action had been taken on the party’s police complaints as far as he knew.

Disinformation researcher Bharat Nayak told Bellingcat that it has always been tech platforms’ responsibility to control new types of content.

“The goal post can’t shift. This has always been a tech problem,” he said.

When this responsibility is shrugged off, Nayak added, the result is a lack of accountability. “If you’re using old videos from other countries as new, you will have people countering you. But AI-generated videos can be shared without context just to spread hate – like showing people in skull caps – and the ‘when, where, how’ questions vanish.”

Both Meta – which owns Facebook and Instagram – and X have policies against hateful conduct.

Meta also announced in 2024 that it would start adding “AI info” labels to more content detected as AI-generated, while some X users spotted a similar feature introduced on the platform last month. Only five of INC’s AI visuals that we identified – and none of those by TMC or the BJP – had a disclaimer that said “AI-generated”.

Bellingcat reached out to Meta and X for comment on whether the posts we identified breached their terms of use regarding hateful conduct or labelling AI-generated posts. A Meta spokesperson said they were reviewing the flagged content and “will take appropriate action on any violations of our policies”. As of publication, X had not responded.

Kalim Ahmed from Bellingcat’s Discord Community contributed research to this piece.