When Tarini Padmanabhuni received the call, her grandfather thought he was talking to his brother. The voice on the other end of the phone said that his brother had been kidnapped and that the only way to rescue him was to pay a ransom. Her grandfather did as instructed, only to later discover that his brother was actually somewhere else and was completely unaware of what was happening. It turned out that that voice was deeply forged—a piece of audio generated by AI that mimicked his brother’s voice.
“What really impressed me wasn’t the money,” Padmanabhuni said when talking about this incident, “but his complete inability to distinguish between them.”
That was about two years ago. Today, she says that the goal of her company DetectifAI in San Francisco is to ensure that no one else will fall for the same trick.
This has already become a significant issue that continues to grow, and the market has correspondingly adapted to it. According to data from FBI, Americans suffered losses of nearly $900 million last year due to scams driven by AI, representing a 24% increase from 2024. The losses incurred by people aged 60 and above were twice those of those aged 50 to 59.
In the field of deepfake voice detection, competition is not uncommon. Relevant companies include Reality Defender, Pindrop, Resemble AI, Microsoft Azure AI Content Safety, and Nuance (Microsoft also owns this company). However, Padmanabhuni believes that today's detection products all run on cloud-based remote servers, which means that smartphone manufacturers are unable to integrate them directly into their devices, leaving those targeted with almost no means of defense.
DetectifAI doesn't involve scaling down large cloud models to fit into mobile phones like some companies do; instead, they design small AI models from the beginning to ensure they can run within smartphone operating systems. The company's goal is to instantly determine whether sounds in calls, voice messages, and other audio are generated by AI, without the need for the audio to leave the device.
DetectifAI will first be sold to mobile phone manufacturers, authorizing their software tools so that the detection function can be launched as a built-in feature of the mobile phone operating system. Padmanabhuni made an analogy: this is similar to the role played by AT &T when the first generation of iPhone was released; at that time, the exclusive cooperation with operators distinguished it from competitors. She believes that the first mobile phone manufacturer to equip with DetectifAI will gain a relative competitive advantage, and deepfake detection will also become a standard feature, just like camera resolution.
The core of this product is the software development toolkit (SDK) of DetectifAI, which is a set of code packages that other companies can integrate into their own products, and it can also be licensed through existing channels. Padmanabhuni indicates that the second source of revenue will come from licensing this technology to enterprises and anti-fraud companies.
Meanwhile, this startup has already generated early revenues. According to Padmanabhuni, it handles over 100,000 calls per month for financial institutions in India. These calls are made by AI voice agents, who are used for debt collection and follow-up on loan documents. Each call is equipped with deepfake detection and speaker verification (to confirm that the caller is indeed who they claim to be). Due to confidentiality agreements, she refused to disclose the names of the clients.
Padmanabhuni said that she started working in machine learning at the age of 12 and later studied Cyber-Physical Systems at Manipal Institute of Technology in India, which is a technology that connects software with physical machinery. She stated that she became the youngest team leader in what she referred to as India's first autonomous racing team there.
When asked about the most fulfilling moment of her entrepreneurial journey to date, she mentioned a small-scale WhatsApp test: users could forward suspicious voice messages and receive an assessment of their authenticity. A tester’s own relative had experienced fraud, and after that, they called to say they were willing to pay for this service without hesitation. “My grandfather didn’t have such a tool,” she said.
Padmanabhuni indicates that DetectifAI has so far raised a small amount of seed funding from investor Josh Constine (formerly edited by TechCrunch) and Manohar Kamath, the person in charge of KM Growth consulting services company. This company is also one of the startups selected by the TechCrunch editorial team and will participate in the Startup Battlefield competition of the TechCrunch Disrupt conference, which will be held from October 13th to 15th in the downtown area of San Francisco.
