Alex Blania brings World ID to Zoom and Cloudflare to fight deepfakes and authenticate AI agents

World for Zoom and the Cloudflare agent protocol

Alex Blania is making a straightforward bet: the deepfake problem is bad enough now that proof-of-human verification is becoming infrastructure, not a nice-to-have.

The first launch is World for Zoom, aimed squarely at financial institutions and public companies. The concern is concrete — a deepfaked CFO on a Zoom call can authorize fraudulent transfers; a deepfaked public company CEO on an earnings call could move markets by billions. OpenExchange, which hosts earnings calls for most public companies, is cited as an early customer with obvious exposure.

The second announcement is an agent protocol partnership with Cloudflare that addresses a different but adjacent problem. As AI agents proliferate, services are blocking them outright — Charles Schwab and JPMorgan have both moved in that direction. Blania's argument is that blanket blocking is a blunt response to a legitimate concern, and World ID offers a more precise one. Rather than proving an agent is human, the protocol lets a service verify two things: that the agent is acting on behalf of a real, unique human, and that a human was in the loop for the specific decision being taken. The practical use cases are things like restaurant bookings and ticketing, where sites are already blocking automated access. A verified human-backed agent could be treated differently than an anonymous bot.

“What happened this week is we launched World for Zoom, especially with financial institutions... The other big one is we announced today that we work on an agent protocol with Cloudflare... a service can ask these two questions: is that agent acting on behalf of a human? And in fact, on behalf of a unique human for that particular service?”

Orb vs. lighter-weight verification

Blania is candid that the orb-based iris scan offers the highest assurance but limits scale. For Zoom, Tools for Humanity is using a phone-based verification track that carries less certainty but is easier to deploy. Users on a call who haven't been orb-verified can still get a credential, just with a different badge to signal the lower assurance level.

On whether software-based deepfake detection could eventually replace the hardware biometric approach, Blania is skeptical. He doesn't dismiss the idea outright but makes clear he thinks relying on video analysis alone isn't sufficient — which is why the orb exists. The counterargument raised in the conversation is that Pangram has maintained reliable AI-text detection for years despite the odds, suggesting detection technology can keep pace even with constrained tooling. Blania doesn't concede the point.

The honest tension here is one of distribution. Proof-of-human is only as useful as its coverage, and right now the orb network isn't everywhere. The phone-based fallback buys scale at the cost of the assurance guarantee that makes the product most valuable to high-stakes users like banks.