A deepfake can fake a face, a voice, a whole resume. Clarity built the machine that calls the bluff - in real time, while the interview is still running.
The candidate is articulate. The camera is steady. The lighting is flattering in the way of all video calls. There is a face, a voice, a name on the application, references that check out. There is, in fact, everything you would expect from a real human being applying for a real job - except a real human being.
What the manager cannot see is the model rendering that face frame by frame, the cloned voice trailing the lips by a few milliseconds, the second person off-camera feeding answers. What the manager cannot see is exactly the problem Clarity was built to solve.
Clarity sits quietly in the corner of that call. It is not watching the answers. It is watching the seams - the micro-stutter where a synthetic mouth doesn't quite close, the eye motion that's a touch too smooth, the audio artifact a human ear glides right past. And somewhere in the dashboard, a flag goes up: this one isn't real.
That is the whole company, distilled. Everything else - the funding, the founders, the patents - is in service of that single, unglamorous moment of doubt.
We are in an AI arms race, and today deepfakes are the new AI cybersecurity threat.
Clarity was founded in 2022 by a team whose backgrounds read less like a startup pitch and more like the opening credits of a spy film. Michael Matias, the CEO, studied AI at Stanford, served as an officer in Israel's 8200 cyber unit, and was named to Goldman Sachs' 2023 Builders and Innovators list. Natalie Fridman, the founding CTO, holds a PhD in computer science with an AI specialization and has published more than 25 papers. Gil Avriel rounds out the founding team after fourteen years as a legal advisor at Israel's National Security Council and a master's from Harvard.
It is an unusual blend - research, intelligence, and policy - but it maps neatly onto the problem. Detecting a fake is part machine learning, part adversarial thinking, part understanding how institutions get fooled. Clarity needed all three in the room.
Stanford AI, officer in the 8200 cyber unit, Goldman Sachs 2023 Builders & Innovators honoree. The public face of the deepfake fight.
PhD in computer science with an AI specialization; author of 25+ research papers. The detection science behind the product.
Harvard master's, fourteen years advising Israel's National Security Council. Knows precisely how systems get manipulated.
Clarity runs an ensemble of models across video, image, and audio, looking for the artifacts a generator leaves behind. Any one signal can be argued with. Stacked together, they're hard to fake all at once.
It installs as a security overlay on the tools companies already use - the applicant tracking system, the video-conferencing platform - so nobody has to change how they hire. It just starts watching.
Spots fake profiles and AI-generated personas before an interview is ever scheduled, using behavioral and identity pattern analysis.
Ensemble AI flags deepfaked faces and voice clones in real time, during the live video interview.
Real-time, AI-powered identity checks woven directly into the hiring workflow.
Catches the case where the person interviewing isn't the person who'll show up to work.
When Clarity raised its $16M seed in February 2024 - led by Walden Catalyst Ventures and Bessemer Venture Partners - the pitch was broad: authenticate media for news outlets, government agencies, intelligence services, journalists. Anyone who needed to know whether a video was real.
It was a good thesis with a hard go-to-market. "Truth" is a large, diffuse customer. So the company followed the pain. And the sharpest pain, it turned out, was in hiring.
Remote work opened a door. Generative AI walked through it. Suddenly fraudsters could manufacture a face, a voice, a portfolio, and an entire interview. State-backed operatives - Clarity's own research documents North Korean IT workers using deepfakes to infiltrate payrolls - turned the job application into an attack surface.
Matias puts the scale bluntly: the company has yet to meet a Fortune 500 that hasn't, somewhere, inadvertently hired a North Korean IT worker. Infiltration attempts, by Clarity's count, climbed 220% in a single year.
So the lab became a product with a clear buyer and a clear moment of truth. The mission narrowed from "authenticate all media" to "make sure the person you're hiring is real" - and got far more sellable in the process.
It's a tidy lesson in startup physics: you can build the most elegant detector in the world, but you ship where the alarm is already ringing.
We have yet to meet a Fortune 500 company that hasn't inadvertently hired a North Korean IT worker.
Matias, Fridman, and Avriel set out to build patent-pending detection across video, image, and audio.
Led by Walden Catalyst Ventures and Bessemer Venture Partners to grow the team and the platform.
Application verification, identity confirmation, and proxy-candidate detection across the employee lifecycle.
Clarity publishes on deepfake-driven remote IT-worker infiltration, citing a 220% surge.
Clarity's customer is the enterprise security and talent team at companies large enough to be a target and remote enough to be exposed - the kind associated with names like AWS Startups, Goldman Sachs, Deloitte, and Intel Ignite. The product is sold the way infrastructure is sold: as a subscription overlay that disappears into the workflow and only speaks up when something is wrong.
That positioning matters. A tool that demands new processes gets quietly ignored by busy recruiters. Clarity's bet is that defense has to be invisible until the moment it isn't - integrated into the applicant tracking system and the video call, flexible APIs ready for whatever stack a company already runs.
It is not alone in the fight. A growing field of detection companies - Reality Defender, Sensity, Hive, voice specialists like Pindrop - is racing the same generators, while identity-verification incumbents bolt deepfake checks onto existing products. Clarity's wedge is the hiring lifecycle specifically: not "is this video real" in the abstract, but "is this hire real" from application through onboarding.
The honest tension running underneath all of it: detection is reactive by nature. Every new generative model is a new puzzle. The work is never finished, which is precisely why a research-heavy founding team - the kind that publishes papers and reads threat reports for fun - is the asset, not the headcount.
The candidate is still articulate. The camera is still steady. Nothing in the room has changed - except that the hiring manager now knows something the candidate hoped they wouldn't. The seams have been counted. The voice has been measured against the lips. The flag is up.
The manager doesn't have to be a forensics expert, doesn't have to squint at pixels or trust a gut feeling. The doubt has been outsourced to a machine that does nothing else all day. The non-existent person is shown the door before they ever get a badge.
Clarity won't claim it has solved deepfakes - nobody honest would, in an arms race that updates weekly. But it has changed one small, decisive moment: the instant a company decides to believe what's on the screen. For now, in that one room, seeing is a little closer to believing again.