Pindrop raises $100M in venture debt to scale voice deepfake protection

Voice deepfake attacks capable of fooling call center agents, automated systems, employees and loved ones have spiked this year, with dramatic increases in the availability and quality of voice-spoofing tools, Pindrop Founder and CEO Vijay Balasubramaniyan told Biometric Update in an interview. This is the context behind $100 million in debt financing secured by Pindrop from Hercules Capital to scale its customer base for deepfake detection and voice biometrics.

The funds will also be used to develop liveness and voice recognition tools for a widening range of industries, as deepfakes are deployed to commit fraud in more scenarios.

“It used to be a cool party trick and now it’s everywhere,” Balasubramaniyan says. He notes that deepfakes followed generative AI into the limelight last year, but this year real-world attacks have taken off.

The combination of large language models and text-to-speech engines allows attackers to scale their synthetic voice fraud efforts based on the number of computers they have at their disposal, not just how many people.

“All of last year, what we were seeing is about an attack every month for our customers,” he says. “It’s now, for the first four months, when we average it out, it’s one attack per customer per day. And there are certain customers who have more attacks every single day.”

That represents a 450 percent increase in the first four months of 2024, compared to all of last year. The rate of increase indicates that figure will be much higher by the end of the year. One in every 730 calls to a contact center by the end of the year will be a fraud attack of some kind, according to the fundraising announcement.

At the same time, the technology is being used to commit more different kinds of fraud, including in election scams like the robocall faking President Joe Biden’s voice that Pindrop unmasked in February.

“I’ve never heard of any technology breaking commerce, media and communications all in one fell swoop,” says Balasubramaniyan. “So that’s the moment we’re in.”

The company had already raised more than $200 million from equity financing, but switched its fundraising approach based on the maturity of its business, Balasubramaniyan explains.

Equity financing operates on a model where one success in ten can pay for the others, whereas “in the venture debt world they can afford very few misses.”

The interest rate, he says, is small compared to massive growth in company value. “It’s a tenth cheaper than the appreciation in equity value and the lack of dilution.”

Pindrop already has the scale, break-even cash flow and positive metrics to avoid dilution, and Balasubramaniyan says the company is clear on its customer acquisition cost.

“If you look at a lot public companies, they operate on debt, because at that point all the numbers are so solid,” he says. “So, it’s just that we’ve matured as a business and this feels like the least-expensive way of getting capital.”

Both scaling Pindrop’s customer base and further product development are planned, and Balasubramaniyan notes a very high rate of its customers adding deepfake detection. “That attach rate is incredible,” he says.

Existing voice biometrics customers can also take the technology live almost immediately.

The level of revenue Pindrop has reached with its deepfake product in 3 months took 3 years for the company’s original voice biometrics product.

John Chambers, a Pindrop investor and board member who is also founder and CEO of JC2 Ventures and a former executive chairman and CEO of Cisco says Pindrop developed its deepfake product exceptionally quickly, immediately secured customers, “and, under Vijay’s leadership, Pindrop will achieve the fastest run rate to $5 million in an entirely new product area by an existing company, compared to any other company I have seen. This new investment from Hercules Capital comes at a pivotal time for Pindrop, enabling them to further develop their cutting-edge audio, voice, and AI technologies and enhance their offerings for customers in the banking, finance, contact center, and insurance industries,” he says in the announcement.

Pindrop also wants to take its voice deepfake detection outside of the contact center “to places like media, government, communication,” Balasubramaniyan says, “so it’s a big TAM expansion opportunity for us.” Over four times bigger, he estimates.

The Biden robocall allowed Pindrop to showcase sophistication level attackers have reached. Pindrop was monitoring a single online tool for voice spoofing when it began working on deepfakes. That figure reached 120 by the end of last year, and by March there were nearly three times as many.

Awareness of the threat has increased with the Biden robocall, but the week-long news cycle in which that fake was revealed is too long to avoid havoc if it occurs again closer to the November election, Balasubramaniyan warns. Progress on public understanding continues, though. “The notion of liveness wasn’t very clear to lots of people. Now we have congressmen talking about liveness,” he notes.

Policy is always catching up to technology, but Balasubramaniyan hopes that it does not become too specific in terms of the technology requirements it imposes. Watermarks, for instance, have a role to play in assuring people that content is genuine. But that role is limited. In the case of the Biden robocall, only 2 percent of the original watermark could be detected, “and that wasn’t even an active adversary, it just went through transmission channels.”

As more content is generated with AI, figuring out whether the subject has consented will become a key challenge.

Perhaps the next being product for Pindrop can be “consent detection.” In the meantime, the company has the resources it needs to address huge and growing market demand for protection against voice deepfakes.

Article Topics

biometric liveness detection  |  biometrics  |  deepfake detection  |  financing  |  investment  |  Pindrop  |  research and development  |  synthetic voice  |  voice biometrics