A Deepfake Scammed Her Grandfather. Now DetectifAI Sells On-Device Detection

Tarini Padmanabhuni's grandfather paid a ransom to a kidnapper who never existed. The voice on the phone claimed to belong to his brother, held hostage and reachable only for a price. The brother was elsewhere the whole time, unaware of the scheme. The caller was a deepfake — an AI-generated imitation of a family member's voice.

"What stayed with me wasn't the money," Padmanabhuni says of the incident, which took place roughly two years ago. "It was that he had no way of telling."

Today she runs DetectifAI, a San Francisco-based startup building deepfake voice detection designed to run directly on smartphones. The company already handles more than 100,000 calls a month for financial institutions in India, where AI voice agents perform debt collections and loan-document follow-ups, with deepfake detection and speaker verification running on every call. Padmanabhuni declined to name those customers, citing confidentiality agreements.

The problem she is attacking is large and growing. According to the FBI, Americans lost close to $900 million to AI-driven scams last year, a 24% increase from 2024. Victims aged 60 and older lost twice as much as those aged 50 to 59. The demographic that scammers target most heavily is also the least equipped to spot synthetic voices.

DetectifAI is not alone in the market. Reality Defender, Pindrop, Resemble AI, Microsoft Azure AI Content Safety, and Nuance (which Microsoft also owns) all offer deepfake voice detection. Padmanabhuni's critique of the incumbents is architectural: their detection products run in the cloud on remote servers, so phone manufacturers cannot build them directly into devices. The person receiving the call, in her telling, is left with no real-time defense.

Rather than shrinking large cloud models to fit a phone — an approach some competitors take — DetectifAI designs compact AI models from scratch, small enough to run inside a smartphone's operating system. The goal is an instant verdict on whether a voice is AI-generated during calls, in voice messages, and in other audio, without the audio ever leaving the device.

The business model

DetectifAI sells first to phone manufacturers, licensing its software development kit so detection can ship as a built-in feature of a phone's operating system. A secondary revenue stream will come from licensing the technology to businesses and fraud-prevention firms, Padmanabhuni says. The SDK can be licensed through existing channels.

Her pitch to handset makers borrows from telecom history. AT&T's exclusive carrier deal at the original iPhone launch set it apart from rivals; similarly, the first phone maker to ship DetectifAI will gain an edge over competitors, she argues. In her view, deepfake detection will become a standard specification, like camera resolution.

The startup has raised a small seed amount from two investors: Josh Constine, formerly an editor at TechCrunch, and Manohar Kamath, a principal at the consulting services firm KM Growth. DetectifAI is one of the startups vetted by TechCrunch's editorial team to compete in its Startup Battlefield competition, held at TechCrunch Disrupt from October 13 to 15 in downtown San Francisco.

From racecars to fraud detection

Padmanabhuni says she began working in machine learning at age 12. She later studied cyber-physical systems — technology linking software with physical machinery — at Manipal Institute of Technology in India, where she became the youngest team lead of what she describes as India's first driverless racecar division in Formula Student, an international student engineering competition.

Asked about the startup's most rewarding moment, she points not to revenue or funding but to a small WhatsApp beta test. Users forward suspicious voice notes and receive an assessment of whether they are real. One tester, whose own relatives had been scammed, called to say they would pay for the product without hesitation.

"My grandfather didn't have that," she says.

The gap DetectifAI is betting on is that incumbents' cloud-based tools cannot reach the moment of deception — the phone call itself. If the company's compact models deliver accurate verdicts on-device, and if a major handset maker licenses them first, deepfake detection could move from niche security product to default phone feature. The competition in San Francisco next month will be an early test of whether the industry buys that thesis.