Zoom adds World's Verified Human to protect video meetings from deepfakes

Zoom has announced a partnership with World, the biometric project linked to Sam Altman, and will add a Verified Human feature to Meetings. The feature is meant to confirm that the person on the other side of a call is a live human, not a generated deepfake. It isn't aimed at routine standups but at sensitive calls where a mistake can cost millions.

How Deep Face works

The integration is built on World ID Deep Face technology. During a check, the system compares three signals: a frame from the live Zoom video stream, a Face Auth selfie check on the participant's device, and the image captured at initial registration through the Orb, World's own biometric scanner. If everything matches, a Verified Human badge appears next to the user's name and video. According to the companies, verification runs locally on the device, and Zoom receives only the result of the check. No personal biometric data is shared with other meeting participants.

• Verification when joining a meeting
• A Deep Face Waiting Room before participants are admitted to the call
• Requests for re-verification during the call
• A Verified Human badge in the meeting interface

This is not a mass-market, set-it-and-forget-it feature, however. To pass the check, a person needs World App and a World ID that has already been verified through a physical Orb device. World has about 18 million verified users in 160 countries and roughly 1,500 active Orbs. Given Zoom's enormous user base, this is for now more of a narrow trust layer for specific scenarios than a universal standard for all work calls.

Why Zoom is moving into biometrics

The reason is simple: deepfake fraud is no longer a rare experiment. In the first quarter of 2025, businesses lost more than $200 million to such attacks. One of the most high-profile cases involved the engineering firm Arup, which lost $25 million in 2024 after a video call in which every other participant, including the supposed CFO, turned out to be an AI double. In 2025, a similar attack hit a multinational company in Singapore.

The market already has tools that try to catch fakes by analyzing the video frames themselves. Zoom and World are betting on a different approach: instead of guessing whether a neural network rendered what's on screen, they confirm identity against a biometric record created in advance. This fits Zoom's strategy as an enterprise platform that needs to sell not just convenience and AI features but also trust for negotiations, payment approvals, executive communications, and other high-risk scenarios.

"Security and trust have always been the foundation of our platform," said

Brendan Ittelson, Zoom's chief ecosystem officer.

Where the risks begin

This approach has a price. World has long faced questions from regulators and privacy advocates over its collection of biometrics, primarily iris images. Authorities in several jurisdictions, including Spain, Germany, and the Philippines, have already raised complaints against the project. World says it uses a privacy-focused architecture, zero-knowledge proofs, and local storage of sensitive data, but for many companies that won't be enough: mandatory biometric registration in itself may already become a legal and reputational problem.

There is also a purely practical barrier. Deep Face is currently in limited beta, and only holders of an Orb-verified World ID can pass the check. That makes the feature too heavy for ordinary meetings with contractors, clients, or job candidates. For treasury teams, M&A negotiations, telemedicine, legal approvals, and other calls where the cost of a mistake is very high, though, this level of friction may well be justified. It's also worth noting that the integration analyzes video specifically, not voice.

What this means

Deepfake protection in video communications is shifting from content detection to proof of identity. Zoom is showing that for businesses, the question "is this really a human?" is becoming a security layer of its own. But mass adoption of such systems will hinge not on model quality but on whether companies and users are willing to accept biometrics as the price of trust.