AI Sexual Abuse Has Arrived in India

In 2025, a woman from the northeastern Indian state of Assam downloaded a loan app called Scoreclimb. Like many Indians, the woman was grappling with rising expenses, and the app’s guarantee of immediate credit gave her respite — or so she thought.

The registration process seemed standard: The woman had to submit her Permanent Account Number (or PAN, a government-issued number used for financial transactions) and upload a selfie taken from her phone as a verification measure, she was instructed.

Before the woman had completed the loan application, ₹1800 (around $19) was deposited into her account. This is a common tactic in fraudulent lending app scams, in which a “token amount” is immediately deposited into the prospective borrower’s account — even before the borrower’s request — so that they are immediately in debt of repayment.

Not long after, the app’s operators began demanding repayment, which the woman promptly returned, hoping it would settle things. Then they demanded more money, citing pending interest. When she refused, things took a sinister turn.

The selfie she had used for her Scoreclimb verification had been uploaded to a nudify app, which then generated a pornographic image of her that included her phone number. The image was then circulated via WhatsApp to unknown groups, as well as to her close contacts.

The woman recruited the help of Meri Trustline, a national helpline that supports victims of online sexual violence with content takedown, counselling, and social and legal support. Trustline advised her to block and report the numbers — a tedious task when unknown callers kept mounting as the image continued to circulate.

It was Trustline that escalated the case to WhatsApp, which disabled the number distributing the content. A follow-up revealed that the woman had stopped receiving threats thereafter.

This story was one of the four AI-related case studies in the report Make It Real: Mapping AI-Facilitated Gendered Harm (2025) published by Rati Foundation — the nonprofit organization behind Meri Trustline — and Tattle Civic Tech, an open source project responding to inaccurate and harmful content in India. The cases were recorded between 2022, when Meri Trustline was founded, and 2025.

Roughly 10% of the cases reported to the Trustline since its founding have involved digitally-manipulated material in which existing media has been altered to harass someone. The report defines image-based sexual abuse as “an act where an individual shares or threatens to share non-consensual intimate images or videos of a person.” Artificial intelligence (AI), in this case, is then defined as “computational systems that are able to generate images, audio, video or text in response to prompts,” which can be used “to create entirely new media or make alterations to existing media.”

While image-based sexual abuse is historically committed by an intimate partner, the Rati Foundation found that AI-generated imagery is also being committed by strangers, who share no prior relationship with the victim at all.

An unexpected perpetrator

The woman told Meri Trustline that, while she’d expected some people to recognize that the image was fake, she still felt as though she had been “involved in something dirty” and was “socially marked.”

Siddharth Pillai, co-founder and director of Rati Foundation — who also co-authored the report — told WMC Women Under Siege that families and communities would not ask whether the images were fake, but instead what the survivor had done for this to happen. “The abuse is designed to contaminate reputation through shame and insinuation,” Pillai said.

In Indian society, even the slightest implication of sexual impropriety can have far-reaching social consequences and reputational damage for women. A national health survey found that, among surveyed women who’d experienced any type of physical or sexual violence, more than 75% never sought any help or told anyone afterward.

That silence is deeply influenced by the social repercussions that often follow abuse in Indian culture: social scrutiny, character assassination, and, in many cases, honor-based violence – which AI-generated sexual abuse weaponizes. The report states that the deepfake wasn’t designed to convince anyone of its veracity but to shame its target, using distortion to render the victim mockable. While the circulated image is a deepfake, the reputational fallout is devastatingly real.

What’s more, the scale of abuse has been amplified with the growth of India’s digital framework: In 2025, the country’s internet user base passed 950 million — driven largely by rapid growth in rural connectivity — with 44% of users using some form of AI‑enabled features, such as image‑based search, chatbots, and AI filters.

“Millions of Indian users came online within a relatively short period, often mobile-first,” said Rohini Lakshané, an interdisciplinary researcher and technologist. “But safety literacy, platform accountability, and legal awareness have not grown at the same pace as connectivity.”

The vast majority of Indians use WhatsApp, which, beyond communication, largely functions as an informal content distribution system. “India has a highly networked forwarding culture,” said Lakshané. “Once manipulated imagery enters a group chat, it can be forwarded with little friction.”

In nearly all Trustline cases, the AI-generated explicit imagery relied on existing material from the victim: photographs that they had published on their profiles.

“Instead of breaking into systems, offenders are increasingly working with what is already available, such as public photographs, social media profiles, and tagged images,” said Gautam S. Mengle, a security awareness strategist for the cybersecurity and data privacy firm Culsight, a phenomenon he called “contextual exploitation.”

When deepfakes first surfaced, creating convincing abusive imagery required technical expertise and specialized tools. But today, Mengle said, “AI has lowered that barrier. A basic set of publicly visible images and a prompt can be enough to create highly convincing abusive material.”

“The abuse was designed with a lesson at its core,” said Pillai. “Visibility comes with a price.”

The failure of platforms to address harm

Meanwhile, said Pillai, moderation systems and platforms have routinely failed to take action until the abuse had escalated enough, and that reason partly lies in platforms’ inability to define harm.

“Social media platforms struggle when abuse does not resemble conventional NCII [non-consensual intimate imagery], such as explicit nudity or violent imagery,” said Pillai, meaning that less explicit imagery — say, AI-generated bikini images — do not violate platform guidelines, thus slipping through the cracks.

“They fail to understand that if the AI bikini image is generated without consent, it is still abusive.”

By Indian law, messaging and social media platforms are required to have a grievance officer whose details must be publicly posted on their website. Under Rule 3(2)(b) of India's IT rules, intermediaries must remove or disable access to content that prima facie depicts nudity, sexual acts, exposes a person’s private parts, or involves impersonation through electronically altered or AI-morphed images within 24 hours of receiving a complaint.

But in reporting AI-generated content to platforms, the Trustline report found that the greatest challenge lies in the overall reporting architecture.

WhatsApp, for example, has a grievance mechanism to review complaints involving morphed, pornographic, vulgar or sexually explicit content shared without consent. But moderators still struggle to determine whether an image is synthetic, edited, satirical, or malicious. “When a platform’s policy requires a clear violation threshold, borderline cases often get labelled as non-violative,” said technologist Lakshané.

What’s more, content moderation decisions are also weighed against liability risk, she said. “Platforms often design policies to minimize legal exposure while preserving user activity.”

Then there’s the issue of their technology: Earlier this year, the social media platform X (formerly Twitter) introduced new guardrails for Grok, its AI chatbot, “to determine whether and under what circumstances the chatbot would generate nonconsensual sexualized images,” Reuters reported. And yet, its reporters found, the chatbot still produced sexualized imagery when prompted — even after being told that the subjects did not consent.

Without responding to the investigation’s findings, xAI instead repeatedly sent a boilerplate response saying, “Legacy Media Lies.”

A lasting impact

For survivors, removing the harmful content seldom means that the harm has ended (much less undone) — especially in cases when the perpetrator is a stranger.

Nirali Bhatia, an Indian cyber psychologist, said that abuse by unknown perpetrators produces a different kind of psychological harm. “When you know the perpetrator, there is awareness of the source of the images,” wherein survivors can anchor the source of harm to a specific decision or relationship.

But when the perpetrator is anonymous, “There is a more persistent form of fear.” And the absence of narrative control — of any identifiable action to undo or correct — intensifies the feelings of powerlessness, she said.

“If you know the person, you are more prepared after the first event. But here, every event is like a first event.”

India has recently introduced a new framework under which AI-generated content that includes non-consensual intimate imagery will be treated as illegal. But gaps emerge when harm is only threatened, said Apar Gupta, a lawyer and founder of the Internet Freedom Foundation. “Harm is not taken seriously when there is only coercion and fear,” he said.

In the absence of any evidence of circulation, and when perpetrators are anonymous or offshore, survivors struggle with seeking quick, coordinated action across platforms. The Scoreclimb case illustrates this loophole.

Deepfake debates in India often center on politics and misinformation, while synthetic sexual abuse gets reduced to bad content instead of being recognized as gendered violence. “The legal system needs to treat it as a systemic safety and dignity issue instead of treating safety reporting as discretionary, inconsistent, and dependent on internal moderation culture,” said Gupta.

And if there is another legal loophole that stands out as stark, it is this: Reporting the content for copyright infringement has proven to be more effective than framing the abuse under gender-based harm. When dealing with platforms, the report stated that reporting NCII as a copyright violation often yielded faster results.

In India, “There are well-established processes and strong liabilities around intellectual property violations,” said Lakshané. By contrast, digital gender-based abuse complaints often require subjective moderation decisions concerning intent, context, manipulation, and potential harm.

“The system is optimized for protecting property rights, not necessarily protecting people from harm.”

For survivors, reframing sexual harm as an intellectual-property issue is often necessary, not because they fit the harm, but because they work. But treating AI-generated sexual abuse as a copyright violation serves only partial justice. It may trigger a takedown, but it fails to address the harassment and humiliation the survivors endure, or hold perpetrators accountable.

As Pillai puts it, the threat has shifted from “I will leak what I have” to “I can generate what I want.”