After a deepfake voice fooled her grandfather, this founder sprang into action

Tarini Padmanabhuni says the whole thing started with a family scam that felt too real to ignore. Her grandfather got a call from what he thought was his brother. The voice said he’d been kidnapped and demanded ransom. He paid. Later, he learned his brother had been nowhere near the scene, and the “caller” was an AI-made clone.

That experience pushed Padmanabhuni toward DetectifAI, the San Francisco startup she founded to catch deepfake voices in the act. Her pitch is simple: if a scam can sound convincing enough to fool an older relative, the defense shouldn’t depend on someone noticing something is off. It should be built into the phone itself.

DetectifAI is betting on a different technical path from many rivals. Instead of taking a big cloud model and squeezing it down, the company says it builds compact models from the start so they can run inside a smartphone’s operating system. That would let the phone make an instant call on whether a voice is AI-generated during live calls, voice messages and other audio, without sending the audio off-device.

The company is selling first to phone manufacturers, with licensing structured so the detection can ship as part of the operating system. Padmanabhuni compares that to AT&T’s role in the original iPhone launch: the first handset maker to ship DetectifAI, she argues, gets a competitive edge, and deepfake detection becomes a normal spec instead of a niche add-on. A second revenue line would come from businesses and fraud-prevention firms.

There’s already some traction behind the idea. Padmanabhuni says DetectifAI has early revenue and handles more than 100,000 calls a month for financial institutions in India, where AI voice agents handle debt collection and loan-document follow-ups. Every call there gets deepfake detection and speaker verification. She declined to name the customers, citing confidentiality agreements. The startup has also raised a small seed amount from Josh Constine and Manohar Kamath, and it’s one of the companies in TechCrunch’s Startup Battlefield competition at Disrupt in San Francisco.

Padmanabhuni, who says she started working in machine learning at 12, also points to a small WhatsApp beta as one of the most satisfying tests so far. Users forward suspicious voice notes and get a verdict back. One tester, whose relatives had already been scammed, said they’d pay for it without hesitation. That, more than any pitch deck slide, is the whole market in one sentence.