Decode at Boom and Tattle Civic Tech partnered on an investigation to understand the infrastructure that enables the use of AI for creating non consensual content. The story published on Decode in June 2026 followed the case of an engineer who created a deepfake influencer ‘Babydoll Archi’ using his ex-girlfriend's face, ran multiple accounts, and made money from them. Unpacking the story revealed layers of an infrastructure that allowed an ex-boyfriend to use a woman's images and create a persona for a decade, without her knowledge. The case epitomises a worrying trend of using advancements in AI to depict women, without their consent, in scenes that have never occurred.

Despite the signalling of intent, and the regulatory interventions by the Indian government, the non-consensually created synthetic imagery (NCSI) continues to grow on the global and Indian internet. The current approach, focused on takedowns, faces a ‘whack-a-mole’ problem where one piece of harmful content is taken down and another resurfaces. In this report we present the more expansive set of interventions that regulators, tech companies as well as individuals need to take across the different layers of NCSI generation infrastructure to be effective in addressing NCSI. Such infrastructural mapping of NCSI moves beyond individual culpability, and also spotlights the accountability of other participants in the infrastructure.

We place NCSI as an outcome of the current paradigm of AI development which is built on ambiguous data access terms, unsustainable resource usage and a unidimensional chase towards approximating human reality. We note that, beyond poisoning their own data, a step that requires significant technical sophistication, there is little that individual users can do to prevent their images from being used in training of models that can produce NCSI, or from being manipulated by the models. Addressing NCSI, in this paradigm, requires regulators, model developers and technology platforms to take multiple significant steps in tandem.

Salient Findings

• Discovering new apps and services is trivial. Apps and websites that can generate nudes or intimate content were found on the first page of all the search engines we checked- Google, DuckDuckGo, Brave. AI chatbots such as Claude and ChatGPT also suggested apps and Telegram bots.
• The infrastructure spans continents. We found websites hosted in Europe, apps hosted in China, API documentation maintained by developers in Bangladesh. Each of these was available to users in India. When monetised, the services relied on intermediaries who converted Unified Payments Interface(UPI) payments from Indian users to cryptocurrency or credit card transactions that the services accepted.
• NCSI generating Telegram bots are widespread and fast growing. Telegram is a common platform to access NCSI-generating services. Many are targeted explicitly at Indians. They offer similar features to websites in Indian languages and take UPI payments. Telegram also provides information for Indians to become agents who can accept UPI payments from other users to convert them into credits required by NCSI-generating bots.
• The Backend Infrastructure is Centralised Many of the Telegram bots were enabled by the same backend or API- bots advertised the ability to clone themselves. New bots and websites names also appeared simultaneously across the Telegram channels, Discord servers and Reddit threads devoted to AI generated content that we were tracking. The coordination suggests centralisation behind the plurality of front-end interfaces.