DetectifAI builds on-device shield against deepfake voice scams

The News

DetectifAI, a San Francisco startup founded by Tarini Padmanabhuni, is building voice-detection models compact enough to run directly on a smartphone, flagging cloned voices during live calls and audio messages without any audio leaving the handset. The company will pitch its technology at Startup Battlefield during TechCrunch Disrupt, held in San Francisco from October 13 to 15, 2026.

The firm has raised a seed round from Josh Constine, a former TechCrunch editor now investing, and Manohar Kamath, a principal at KM Growth. It is already earning revenue and, notably, processes more than 100,000 calls a month for financial institutions in India.

Padmanabhuni founded the company after a personal shock: her grandfather was scammed by a deepfake voice impersonating his brother and demanding a ransom. "What stayed with me wasn't the money," she said. "It was that he had no way of telling." Rather than shrinking bulky cloud models, DetectifAI designs its detection systems from scratch for phone operating systems, then licenses the software kit to handset makers, with additional income from businesses and fraud-prevention companies.

Why It Matters

Voice cloning has crossed the line from novelty to weapon. The FBI reported that Americans lost 900 million dollars to AI-driven scams in 2025, a rise of 24 per cent on 2024, with adults aged 60 and over losing roughly twice as much as those in their fifties. The victims are rarely careless; they simply have no reliable way to tell a real voice from a synthetic one.

That is the gap DetectifAI is chasing, and its architectural choice matters. Keeping detection on the device sidesteps the privacy and latency problems that plague cloud-based fraud tools, a design philosophy that echoes the wider industry shift towards on-device intelligence seen since Apple and Google began pushing neural processing into consumer phones. When fraud detection has to happen inside the few seconds of a live call, milliseconds and privacy are not luxuries but the whole product.

The founder's route is unusual, too. Padmanabhuni began working in machine learning at 12, studied cyber-physical systems at the Manipal Institute of Technology, and led India's first driverless racecar team in Formula Student before turning to fraud detection.

Indian Angle

India is not a footnote in this story; it is the proving ground. DetectifAI already handles over 100,000 calls a month for Indian financial institutions, a scale that reflects how acute voice fraud has become here. The Reserve Bank of India and the Home Ministry's cyber-crime wing have repeatedly warned about "digital arrest" scams and voice-cloning frauds targeting elderly account holders, and banks are under pressure to build defences that work on the low-cost Android handsets most Indians actually use.

That makes an on-device, SDK-first model unusually well suited to India. Cloud-based voice analysis is expensive to run at rupee-sensitive price points and awkward under tightening data-localisation expectations; a model that runs locally on a mid-range phone fits both the cost curve and the regulatory mood shaped by the Digital Personal Data Protection Act.

The talent thread is just as telling. A Manipal-trained engineer building fraud-detection infrastructure for Indian banks from San Francisco underlines how Indian institutions have become early, demanding customers for frontier safety tools, and how Indian-origin founders are turning that domestic fraud problem into a global product.

FAQ

What exactly does DetectifAI detect?

It flags synthetic or cloned voices during live phone calls and in audio messages, alerting the user in real time. The analysis runs on the phone itself, so the audio is not uploaded to a server, which protects privacy and cuts the delay that would otherwise make live warnings useless.

How is it different from cloud fraud tools?

Most fraud systems shrink large cloud models to fit smaller devices. DetectifAI instead builds compact models from the ground up for smartphone operating systems, then licenses them to phone manufacturers and businesses, keeping detection local and fast.

Why does India matter to the company?

India is already a major market: DetectifAI processes more than 100,000 calls a month for Indian financial institutions, making the country central to both its revenue and its real-world testing.

Where can I read the original announcement?

The full profile was published by TechCrunch, linked below.

This story was reported by TechCrunch. Read the full original coverage at TechCrunch.