This Startup Wants To Make Deepfake Detection As Easy As An API Call

Who could be a better guarantee for your money in India than Infosys cofounder Narayana Murthy or Reliance chairman Mukesh Ambani? Two individuals bit the bait of fake images that eventually led to an INR 95 Lakh fraud. While someday a Mumbai-based doctor is duped INR 7 Lakh in a share market scam, a Bengaluru chartered accountant is conned into an INR 23 Lakh trickery on some other day. Such stories hardly surprise us any more as newspapers and tubes keep flashing such incidents almost every day.  

What makes a consumer trust a product? A credible face, beyond doubt. Thanks to artificial intelligence, bringing faces into the picture has never been easier. Scammers have leveraged deepfake technology to feature business icons like Murthy and Ambani in fake trading apps, driving people into the trap.

Not the business tycoons alone, a few months back, a deepfake video featured actor Rashmika Mandanna’s face digitally altered onto British-Indian social media figure Zara Patel. The video gained significant traction due to Rashmika’s popularity.

The Delhi High Court recently directed a subcommittee under the Ministry of Electronics and Information Technology (MeitY) to submit its report on deepfakes by July 21. In the petition, journalist and India TV editor-in-chief Rajat Sharma sought regulations on deepfakes and directives to block public access to apps that facilitate the creation of such synthetic content.

Deepfakes, an emergent type of threat falling under the greater and more pervasive umbrella of synthetic media, use a form of AI or ML to create believable, realistic videos, pictures, audio, and text of events which never happened, says a US Homeland Security communication. 

From face-swapping and voice synthesis to gesture and body movement manipulation, deepfake scams are evolving rapidly. Hybrid deepfakes, which combine multiple AI techniques, are making fraudsters even more powerful. On the other hand, AI-powered deepfake detection tools are also emerging to combat the threat.

It seems to be the return of Frankenstein in this age of artificial intelligence. 

As deepfakes take a $500,000 toll on businesses on the average and burn a $680,000 hole in the pockets of large enterprises around the world in 2024, India stares at INR 70,000 Cr deepfake frauds this year, surging 550% since 2019. On the back of this simmering threat, a battle is brewing up on the country’s digital turf, where AI is being used to take on the menace unleashed by itself. 

Neural Defend entered the battlefront to take this cerebral war to the next level. Founded in 2024 by Piyush Verma, Sivashankar Selvarajan, and Sumit Singh, the Delhi NCR-based startupfocusses on developing AI-powered detection tools that could identify deepfake manipulations across images, videos, and audio.

A Shared Vision: Battling AI-Driven Identity Theft

Neural Defend was born out of a vision shared by the founder trio. Selvarajan, Singh and Verma recognised the growing threat of AI-generated deepfakes and identity fraud.

Selvarajan is seasoned in the AI domain across industries, starting his career with USD Global as an AI software developer working primarily for Intel, in early 2020. This was when he met Singh, who later introduced him to Verma. The three quickly realised that they shared a similar mindset, particularly when it came to the risks associated with AI-generated content.

As AI technology advanced, they saw a much bigger problem unfolding that was far beyond financial fraud. “Identity theft by the use of generative AI is emerging as one of the biggest threats for the future, perhaps second only to climate change,” said Selvarajan.

People are losing control over their own identities, with deepfake technology making it easier than ever to manipulate images, videos, and audio in dangerously realistic ways. In 2023, deepfake-driven face swaps used to bypass identity verification surged by 704%. AI-powered language models have made phishing scams more common and convincing, manipulating people into revealing sensitive information at an unprecedented scale, says a World Economic Forum report. It estimates global cybercrime at $10.5 Tn annually by 2025, up from $3 Tn in 2015.

By the use of AI and spoofing software, scammers have been creating fake police stations and even fake court hearings or using the voice of relatives to dupe victims. A textile tycoon was defrauded $830,000 last year after fraudsters summoned him to an online hearing at a fake Supreme Court that threatened him with jail.

The techie trio noted that only a handful of companies were tackling deepfake detection, and even those solutions were fragmented. AI-generated content was evolving rapidly, but there was no universal mechanism to detect it effectively.

The realisation led them to a decisive moment. Instead of just observing the problem, they wanted to build a solution. This is where the idea for Neural Defend took shape. “Our goal was not just to create another AI business but to restore trust in digital content and provide people with a way to safeguard their identities,” Selvarajan said.

“The name Neural Defend reflects our goal. Neural represents AI and neural networks, while Defend underscores our commitment to fighting against the misuse of AI,” he added.

The founders carried out extensive market research to zero in on the problem. They analysed various industries where deepfake detection was crucial such as dating apps, telecom, call centres, and video verification. Discussions with experts revealed that deepfake threats were rampant in eKYC (electronic Know Your Customer) providers.

They found that over 150 Mn eKYC verifications were conducted every year. This data point validated their concerns and even expanded their initial problem statement, highlighting the need for deepfake detection across images, videos, and audio.

The Proprietary Model: Preventing Deepfake Manipulation

Neural Defend has developed its own proprietary algorithm, avoiding open-source dependencies and securing four patents so far. To ensure seamless integration, it designed the solution as an API, rather than a standalone application.

This API-based approach allows eKYC verification companies to strengthen their existing systems without major code modifications. Selvarajan said clients can simply send an image, video, or audio file and receive an instant deepfake detection result.

The solution focuses on deepfake detection across image, audio, and video verification. The startup is targeting fintech companies, small finance banks, and eKYC verification providers to enhance security in digital identity verification. Neural Defend is also working with Fortune 500 companies and major video conferencing platforms to address security risks associated with deepfake-based identity fraud in virtual meetings.

Financial institutions and digital verification providers use video-based authentication, but deepfake advancements allow fraudsters to impersonate real individuals. The API integrates into video verification systems, analyse each frame and audio signal in real time to detect deepfakes, face masks, and synthetic voices, ensuring secure identity verification.

For video conferencing securities, it offers a plugin for platforms like Zoom and Google Meet. Similar to AI note-takers, it continuously analyses video and audio feeds to verify user authenticity and detect AI-generated intrusions.

Online interviews are increasingly vulnerable to deepfake-based impersonation, where candidates use lip-syncing or AI-generated avatars to cheat the process. Neural Defend’s solution also helps hiring platforms detect such manipulations, ensuring only genuine candidates participate.

The startup operates on a usage-based pricing model, charging per API call for verification services and per minute of video conferencing usage.

The Goal: Touching The $10 Mn ARR Landmark

The market for AI deepfake detector tools is projected to grow significantly from  $1.3 Bn in 2024 to $4.1 Bn by 2032, reflecting a CAGR of 15.1%. The threat is affirmed by the McAfee’s State of the Scamiverse report, which shows how deepfake scams can be created for just $5 (less than INR 430) in under 10 minutes.  

The looming threat and escalating demand for solutions appear promising for businesses like Neural Defend. The startup recently raised over $600,000 in a pre-seed round led by Inflection Point Ventures (IPV). According to the cofounder, this capital will sustain operations for the near future. Riding on the booming demand, Neural Defend is preparing for a seed round at a higher valuation.

Over the next 12 months, Neural Defend aims to secure at least 10 customers. “Given the scale of verification needs, where a single company can conduct 150 Mn API verifications a day, even acquiring two clients could make the company profitable,” Selvarajan said.

The startup aims to reach $10 Mn in annual recurring revenue (ARR) within a year. It also plans to expand globally by tailoring its deepfake detection solutions for specific industries and ensuring seamless adoption across markets.

Although the opportunity for such startups is indeed huge, a challenge is emerging with cybersecurity giants like McAfee rolling out the AI-powered McAfee Deepfake Detector service in India. As the AI-versus-AI duel sets off a storm in India, it remains to be seen how small startups like Neural Defend brainstorm over the way to success.

[Edited By Kumar Chatterjee]