Smaller AI models likely powering India's escalating deepfake menace

Cheap, customised AI tools are making deepfake scams harder to detect, exposing banks, fintechs and consumers to rising fraud risks

India’s escalating deepfake threat is being fuelled less by frontier artificial intelligence (AI) models and more by customised applications built on open-source models, enabling fraudsters to create increasingly sophisticated scams at a much lower cost. 

Deepfakes are becoming so convincing that they can scarcely be distinguished from genuine images and videos, raising fears that fraudsters could use synthetic identities and manipulate live Know Your Customer (KYC) checks to slip through defences designed to prevent impersonation. 

An industry source, who requested anonymity, said the threat had begun to permeate the banking, financial services and insurance (BFSI) sector. 

“People are generating synthetic bank statements and there are issues of deepfakes during video KYC. There have been cases where such applications have been processed and underwritten leading to a fraud of ₹15 to 20 crore at an NBFC (non-banking financial company),” the person said. 

But, the cost of organised fraud is getting increasingly affordable to fraudsters and likewise highly cumbersome

to the industry, thanks to the rapid pace of development of AI models. 

Executives said the frauds typically follow a familiar pattern: customised tools built on top of these models are distributed through Telegram channels and dark-web marketplaces. 

“The challenge is that fraudsters are not using the standard models. Frontier models use trillions of tokens to train and are sophisticated. Instead, fraudsters use smaller models to generate a deepfake and run them on a local, consumer hardware like a gaming computer. It does not cost much. These models are shared across Telegram channels and other dark web platforms,” said Sandesh Gs, chief technology officer (CTO), Bureau; a technology service provider (TSP). 

Deepfake injection involves injecting images into a compromised device that could defeat liveliness checks, while they also include instances of AI-based document tampering. 

These deepfakes pass have the ability to pass most liveliness checks especially at a time when KYC is designed around the principle that a live human face on camera is the proof of presence of the person. 

The rapid proliferation of such tools among organised fraud networks is likely to have prompted the Indian Cyber Crime Coordination Centre (I4C) to issue an advisory warning that fraudsters were using AI-powered techniques to circumvent existing cybersecurity safeguards. 

Industry executives explained that some of the fraud models were specifically trained for deepfakes and could be built on open source models. 

“These are wrappers built on top of an open source LLM (large language model), which is easily available at a fairly cheap cost. Also, the expertise required to do all these sophisticated attacks has come down significantly,” said Prakarsh Paritosh, Principal Product Manager (PM), IDfy; a TSP. 

He explained that the company was working on rolling out deepfake detection modules that could have

the ability to detect synthetic identities after analysing an image. 

 

Banks in the country reported fewer fraud cases in the financial year 2025-26 (FY26), but the total amount involved rose to its highest in three years, driven by cheating in loans and advances and concentrated in state-owned lenders. 

The industry reported frauds worth ₹48,021 crore in FY26, up 46.4 per cent from ₹32,803 crore in FY25, according to data released by the Reserve Bank of India (RBI) in its latest annual report. 

“Deepfakes are extremely hard to detect to a normal person. The reason why this issue is also a growing concern is because devices themselves are compromised. Fraudsters can inject images or deepfakes in that case. First priority should be securing the devices,” Sandesh from Bureau said. 

These frauds do not stop just at the BFSI sector. 

These are a growing threat to e-commerce and quick commerce platforms, social media sites, dating apps, among other things.

“There are sophisticated mule networks who carry on all such frauds. Even the frontier models around identity accounts such as creating a fake cohort to exploit vulnerabilities, is a possibility. Especially for documentation, where some documents are generated by ChatGPT or Claude,” Prakarsh from IDfy said.

 

Executives added that since most of these deepfakes are generated using models trained for malicious intent, they bypass GAN (Generative Adversarial Network) fingerprints or synthetic IDs.

 

These are unique identifiers which result in invisible patterns left behind images or data to detect deepfakes.

 

The I4C has also recommended that ‘customer onboarding systems, including fintech companies should integrate deep fake and synthetically generated content detection mechanisms’.

 

It has called on users to report suspicious activity and identity theft to national cybercrime reporting portal (cybercrime.gov.in).