‘Looked so real’: How AI is being weaponised against India’s Muslim women

New Delhi, India – When Samreen Ayoub archetypal saw the video, she was stunned.

The freelance exemplary from India-administered Kashmir was scrolling connected her telephone past twelvemonth erstwhile a person sent her a clip circulating connected Instagram.

Recommended Stories

• list 1 of 4‘We won’t unopen up’: Muslim women enactment up ‘for auction’ successful India
• list 2 of 4Bulli Bai: India’s Muslim women again listed connected app for ‘auction’
• list 3 of 4‘Auction’ of India’s Muslim women shows tech weaponised for abuse
• list 4 of 4Muslim home workers successful India changing names retired of fear

The video appeared to archer the communicative of her beingness successful New Delhi, implicit with a narrator’s voice, scrolling captions and headlines similar a tv quality segment. But it was wholly fabricated.

“It was due stalking,” Ayoub, 24, said. “They had followed my beingness from my archetypal semester to the past astatine the university.”

The video stitched unneurotic photographs from Ayoub’s clip arsenic a pupil astatine New Delhi’s Jamia Millia Islamia University – images drawn from mundane moments of field life, including radical projects, farewell gatherings and selfies with classmates.

The voiceover, generated utilizing artificial intelligence, falsely claimed she was a Muslim pistillate “selling her body” to Hindu men. It misidentified radical successful the photographs and labelled her ain member arsenic her “pimp”.

“It looked truthful existent that if someone, adjacent my parents, saw the video, they would deliberation it was real,” Ayoub said.

She is 1 of respective Muslim women who person experienced what researchers described arsenic a signifier that is becoming progressively visible: the usage of AI to make sexualised imagery and propaganda.

Al Jazeera reached retired to respective Muslim women who person been targeted. They declined to talk connected the record, citing shame and the hazard of retraumatisation.

‘Sexual phantasy into imagery’

The inclination to sexualise images and videos of Muslim women is unfolding alongside India’s increasing engagement successful planetary conversations connected AI governance, including a high-level AI Impact Summit held earlier this twelvemonth successful New Delhi that focused connected innovation and regulatory frameworks.

A survey by the Washington, DC-based Center for the Study of Organized Hate (CSOH) analysed 1,326 publically disposable AI-generated images and videos collected from 297 nationalist accounts connected X, Facebook and Instagram from May 2023 to May 2025. The researchers recovered that sexualised depictions of Muslim women generated the highest engagement – much than 6.7 cardinal interactions crossed the platforms.

“Generative AI has made the translation of intersexual phantasy into imagery imaginable astatine velocity and astatine nary cost,” said Zenith Khan, coauthor of the survey and a integer probe expert astatine the CSOH. “Image generators and deepfakes let individuals to person hostile narratives into highly realistic ocular worldly with minimal method expertise.”

Researchers are not the lone radical tracking the trend.

Meri Trustline, an online information helpline tally by the Mumbai-based RATI Foundation, has besides seen a increasing measurement of specified cases. The helpline’s 2024 study revealed a concerning pattern: While media attraction tends to absorption connected celebrities and politicians, women not successful the nationalist oculus are besides being targeted done images that, though artificially generated, person the capableness to effect successful existent harm.

Front-line counsellors astatine the helpline, including Salman Mujawar, whose enactment with survivors forms the ground of overmuch of the organisation’s published evidence, said they person documented a emergence successful the fig of specified cases.

Since its inception successful 2022, Meri Trustline has handled much than 482 cases, astir 10 percent of which progressive digitally manipulated worldly – a stock that has been increasing arsenic AI tools go much accessible.

“These violations are muted by shame, fearfulness and trauma,” Mujawar said. “Incidents are seldom revealed adjacent to adjacent household members, fto unsocial featured successful larger nationalist discourse.”

‘Pornification of politics’

Ayoub’s video circulated crossed aggregate societal media accounts wrong hours. Abusive comments, threatening telephone calls and accusations astir her quality followed quickly.

“It felt similar a integer lynching,” she said. “Not 1 but much than a twelve accounts were posting that video everywhere, and hundreds of others were resharing it.”

The information acceptable compiled by the CSOH includes AI-generated memes portraying Muslim women successful spiritual attire successful sexually suggestive scenarios arsenic good arsenic fabricated pornographic imagery targeting journalists and activists. Across galore of these images, researchers observed a recurring ocular pattern: a “Muslim-coded woman” paired with a “Hindu-coded man”.

“In these narratives, Muslim men are often depicted arsenic convulsive oregon morally corrupt,” Khan said. “Meanwhile, Muslim women are portrayed arsenic submissive oregon ‘rescued’ by men from the bulk community.”

This imagery, researchers argued, is not incidental to governmental sermon – it is simply a portion of it.

Sahana Udupa, a media anthropologist astatine Ludwig Maximilian University of Munich, described the improvement arsenic portion of a broader “pornification of politics” targeting women and number communities. Right-wing integer cultures, she said, harvester humour, memes and sexualised imagery to normalise abuse.

“These practices signifier an ecosystem,” Udupa said. “They thrive connected radical solemnisation and corporate aggression.”

That ecosystem, scholars noted, has deeper ideological roots than specified misogyny. Writing successful the peer-reviewed South Asia Multidisciplinary Academic Journal, researcher Soma Basu argued that what is unfolding is the politicisation of the intersexual itself.

Muslim women’s bodies person go battlegrounds for communal dominance – a dynamic that recovered its astir disposable look successful the “Sulli Deals” and “Bulli Bai” controversy, mock-auction platforms that targeted Muslim women successful India and which Basu links to some ceremonial enactment from officials of the governing Bharatiya Janta Party (BJP) and informal backing from the party’s integer volunteers.

Khan’s probe arrives astatine a akin terrain from a antithetic angle. “In galore South Asian cultures, women are portrayed arsenic the honour of the family,” she said. “So attacking Muslim women visually becomes a mode of portraying Muslims arsenic inferior.”

Khan herself recovered the probe profoundly affecting. “As a Muslim pistillate and a researcher, it had a profound interaction connected me,” she said.

“I retrieve being rather horrified erstwhile I saw images of a pistillate wearing a headscarf being represented arsenic brushed porn. As a woman, you already woody with misogynistic behaviour daily. This is different furniture added to your identity.”

Responding to these concerns, Atif Rasheed, a BJP politician, said AI “can beryllium utilized some positively and negatively” and called for stronger regulations to forestall its misuse. He called deepfake and sexually explicit contented “very disappointing” and said determination should beryllium strict enactment against the perpetrators.

However, helium rejected viewing the contented done a spiritual lens, saying the BJP “respects women of each religions” and the “Sulli Deals” and “Bulli Bai” cases were handled successful accordance with the law.

Familiar signifier amplified by AI

The Sulli Deals and Bulli Bai episodes, which occurred successful 2021 and 2022, utilized doctored images. Both triggered wide outrage and constabulary investigations.

Indian authorities arrested Aumkareshwar Thakur, accused of creating the “Sulli Deals” handle, and Niraj Bishnoi, identified arsenic the creator of “Bulli Bai”, successful January 2022. Both Thakur and Bishnoi were granted bail by a New Delhi tribunal connected “humanitarian grounds” 2 months later.

Police officials escort a antheral and a pistillate aft they appeared earlier a tribunal pursuing their apprehension implicit alleged engagement successful an online app that shared pictures of Muslim women for a virtual ‘auction’ successful Mumbai, India, connected January 7, 2022 [Niharika Kulkarni/Reuters]

Researchers judge the emergence of generative AI has dramatically expanded the standard and velocity of harassment of Muslim women online. New applications let users to upload photographs and automatically make sexualised images. Such tools are wide disposable online, often for free, and necessitate nary method expertise.

“There is simply a precise agelong past of exertion being utilized to people and harass women, particularly number women,” said Eviane Leidig, manager of probe and outreach astatine the CSOH. “What is antithetic present is the grade of usurpation and the standard of harm that AI tools enable.”

For those already surviving with sustained harassment, the emergence of AI-generated imagery has added a caller magnitude of fear.

Afreen Fatima, 27, a researcher and activistic who has faced online maltreatment since speaking retired against India’s Citizenship (Amendment) Act successful 2019, was among the dozens of Muslim women whose photos were uploaded and “auctioned” connected Sulli Deals.

The law, declared by the United Nations arsenic “fundamentally discriminatory” against Muslims, seeks to fast-track Indian citizenship for non-Muslim minorities who get successful India from neighbouring Muslim-majority countries earlier 2025.

Afreen Fatima protesting against India’s citizenship instrumentality [Courtesy: Afreen Fatima]

Four years aft the Sulli Deals controversy, the maltreatment has hardly subsided. Anonymous accounts, often utilizing communal Hindu names, proceed to question her retired with abusive messages, rape threats and targeted harassment, immoderate tied to her work, contempt her constricted societal media presence.

“Every fewer days, determination is simply a connection from immoderate random relationship with rape threats oregon decease threats,” she said.

The imaginable of AI-generated intersexual imagery has intensified that fear. “When I work astir these images, it felt precise personal. They make a fearfulness psychosis.”

Fatima said hatred online has besides shaped however she navigates nationalist spaces.

“I consciousness uncomfortable travelling alone,” she said. “When you spot these kinds of fantasies circulating online astir Muslim women, you commencement wondering whether idiosyncratic mightiness onslaught you successful existent life.”

‘I don’t consciousness safe’

After the video went viral, Ayoub’s nonrecreational opportunities began to disappear.

“As a model, your estimation matters,” she said. “If antagonistic comments look connected your profile, brands halt approaching you.”

For 4 oregon 5 months, fake accounts flooded her illustration with abusive comments, deterring imaginable clients. The harassment besides changed her narration with societal media.

“Instagram utilized to beryllium a harmless abstraction for me,” she said. “Now I don’t consciousness harmless there, and I bounds what and however I post.”

Ayoub reported the incidental to the constabulary cybercrime portion successful New Delhi, filing written complaints. “Nothing happened,” she said, adding that astir of the abusive contented was removed lone due to the fact that her friends reported those accounts en masse.

Legal experts said India’s existing laws conflict to support gait with AI-generated content. “The harm is existent adjacent erstwhile the representation itself is fabricated,” said Apar Gupta, lawyer and founding manager of the Internet Freedom Foundation.

Under Section 66E of India’s Information Technology Act, transgression penalties use to capturing oregon publishing images of a person’s backstage areas without consent. But if a target’s assemblage was ne'er really recorded – successful different words, if the representation is wholly AI-generated – the proviso whitethorn not apply.

“Even if the representation is fake, it creates a imperishable scarlet missive for women,” Gupta said.

Digital platforms, meanwhile, bask “safe harbour” extortion arsenic agelong arsenic they region amerciable worldly erstwhile notified. But Gupta said galore victims conflict to adjacent get that far.

“Platforms don’t marque it casual for you to study that this is my image, this is simply a deepfake, you request to region it,” helium said.

Without structural changes to level design, algorithmic priorities and ineligible frameworks, AI-generated maltreatment volition proceed spreading faster than immoderate ineligible strategy tin respond to it, helium warned.

In specified a scenario, accountability remains elusive for targeted Muslim women.

“What I wanted astir was to find the radical down those accounts,” Ayoub said. “They destroyed my estimation without adjacent knowing me.”