Siri co-founder Dag Kittlaus backs Helix, a startup using voice to prove who is human online

Siri co-founder Dag Kittlaus has invested in Helix, a startup that uses on-device voice analysis to verify whether an internet user is human, and will serve as an advisor to the company. Helix raised $6 million in a round led by Chicago Ventures, with participation from BAM Ventures, Aglaé Ventures, Corazon Capital, Tusk Ventures, Yonder Ventures, Rule30 and Brown Alumni Group.

The bet lands at a moment when the internet's oldest identity problem is getting harder. Websites have relied on CAPTCHA for years to separate people from bots. Social media platforms have more recently turned to photos and videos to verify identities. Helix argues that voice can do the same job without forcing users to hand over sensitive personal information, and that the spread of AI makes the problem urgent enough to justify a new approach.

Who founded Helix, and why?

Alex Oberg, Anna Dorofiyenko and Mike Hall founded Helix in 2025. Oberg previously worked at Ticketmaster, which acquired his startup; at Ticketmaster he played a role in creating Verified Fan, the system designed to keep bots off the platform. Dorofiyenko led data science teams at Live Nation and Zillow. Hall has worked at Nike and Tinder, where he helped develop the initial matching algorithm.

Oberg said the idea took shape as generative AI advanced over the past two years. "As AI started to come up in the last two years, I was realizing that it's going to make it increasingly difficult to determine who's human and who's not on the Internet," he said. "Identity on the Internet's always been about proxies for identity, right? It's your email, it's your phone, it's your IP address, it's your device fingerprint, it's your device ID."

His argument is that those proxies — email addresses, phone numbers, IP addresses, device fingerprints, device IDs — no longer settle the question of who is human when AI agents can present the same signals. He pointed to the rise of AI agents like Instinct and Muse as a reason the industry needs to rethink human identity online and rely instead on biometric signals such as voice. Helix is the company built around that premise.

What does the technology actually do?

Helix's core system runs directly on a user's device. It determines whether the speaker is human and estimates their age. The company is also running pilots to explore whether voice can serve as a biomarker to detect conditions such as drunkenness or distress.

The company positions the technology as a replacement for, or supplement to, CAPTCHA, and as a tool to block bot attacks. Because the model runs locally — Helix says it works even on an iPhone 13 — the cost of deploying it is minimal, according to the company.

One customer is already live. Zeel, an in-home healthcare provider, uses Helix's technology to verify the identities of providers visiting customers' homes. Oberg said verifying workers such as cab drivers is another use case the startup is exploring.

The stakes extend beyond bot defense. Age-verification regulations for social media platforms are spreading around the world, and Oberg said those rules are forcing users to trade privacy for access. He frames Helix's four-second voice challenge as an alternative to document-based or face-scan verification.

"There's like an avalanche of regulation coming down because essentially society has decided children should not be able to use the Internet as adults," Oberg said. "That has created a bit of a strawman argument that everyone needs to trade their privacy for the sake of security and child safety. We are that company that's saying, nope, don't agree. It's a four-second voice challenge."

What are the technology's limits?

Helix is candid about what its system cannot do. Oberg said the technology cannot reliably detect whether someone is lying. He also said it does not identify individual speakers, and he claimed that even if the company stores a voiceprint, it cannot be used to reconstruct a person's voice.

AI-powered fraud on phone calls is a growing problem, but Helix acknowledges its technology is not yet capable of reliably detecting AI agents during live calls. That limitation matters, because real-time voice fraud is one of the most visible forms of AI abuse, and Helix's current deployments sit on the verification side rather than the live-detection side.

The privacy claims rest on the company's own statements. Helix says local, on-device processing means sensitive audio does not need to travel to a server, and that the stored voiceprint cannot be reversed into a copy of the speaker's voice. Those claims will face scrutiny as the company scales.

Why did Dag Kittlaus invest?

Kittlaus, who co-founded Siri, said voice identification technology has finally reached the point where it works reliably, which positions Helix to scale as AI-powered fraud increases. He also cited Helix's work compressing its models as a source of cost advantage — a point that aligns with the company's claim that its system runs on hardware as modest as an iPhone 13.

Kittlaus drew a distinction between speech recognition and what Helix does. "What folks at Helix are doing is, instead of teaching computers how to understand speech, these guys are understanding, what the speech of the speaker actually means, what's going on with them, who they are they lying, are they impaired?" he told TechCrunch. "That side of it is fascinating to me within the larger picture that I just described."

The distinction matters for how Helix competes. Voice assistants spent a decade teaching machines to transcribe words. Helix is selling inference about the speaker — human or not, adult or minor, impaired or not — as a service that websites, platforms and marketplaces can drop into their signup, login and worker-verification flows.

What comes next?

The $6 million round gives Helix runway to move beyond pilots. The company's near-term markets are visible in its existing work: CAPTCHA replacement and bot blocking, age estimation under tightening social media regulation, and identity checks for workers who enter other people's homes or cars. Its biomarker pilots on drunkenness and distress, and the unsolved problem of detecting AI agents on live calls, mark the farther edge of the roadmap. Whether regulators accept a four-second voice challenge as sufficient proof of age — the use case Oberg calls an avalanche — will likely decide how large this market becomes.