An Internet With More Humanity

At the 2026 Web Summit in Rio de Janeiro, the World Economic Forum invited me to join a discussion on how AI can improve productivity in Latin America. We covered the Brazilian government's investments in AI, Rio de Janeiro's local hardware infrastructure, the role of universities in building human capital and technology, and initiatives to train large language models on Portuguese and Spanish datasets. There's a long road ahead for our countries to compete in this landscape, but I'm glad we are making strong strides toward innovation.

Still, focusing on productivity, I missed one important aspect of the discussion, one that touches on my role at Unico: the security and humanity of AI agents. The two might not sound related at first glance, but they are more intertwined than they appear. Cloudflare's CEO recently posted on X that bots now outnumber humans in web traffic. The internet, the tool that revolutionized human communication, increasingly belongs to machines, and that reality will only become more common. Take me as an example: this year, I used Claude Desktop to fill out my income tax declaration, and with no free tables at restaurants in Rio de Janeiro on Brazilian Valentine's Day, I asked OpenClaw to find a nice place for June 12th. Both saved me a lot of time, and they are exactly the use cases the World Economic Forum is looking for.

On the security front, Unico saw a 1,000% increase in what we call sophisticated liveness spoofing in 2025. In 2026, we are blocking more than 200,000 new attack attempts per month. These are the attacks that worry everyone today: AI-generated deepfakes. Fraudsters take the same models everyone uses to create realistic Marvel or Star Wars scenes and turn them against financial institutions. As someone who has spent the last 15 years building anti-fraud technologies, I can say the world is upside down for security folks, too. At Unico, we reacted quickly and successfully: we trained a model on tens of millions of sophisticated spoofing samples, implemented additional device security measures, and built robust time-series models that detect attacks before they occur. It proves that liveness detection takes more than a strong model on a server. Like every other anti-fraud discipline, it requires a strong operation.

I joined Unico four years ago with the conviction that my experience building large-scale anti-abuse infrastructure at Google could help protect the Brazilian market. Abuse and fraud are my business, and building a strong identity system is how Unico fights them. But in this new world dominated by bots, I believe anti-fraud and security are only the tip of a larger iceberg. We cannot deny the existence of agents acting on our behalf. The question that remains is: how do we know a human is accountable for that action? My agents act with my permission; the danger starts when a bot claims a permission it never received. Proving that a human exists behind an account is the easier part of the problem. Binding a specific action to that human's consent is much harder, and solving it is where identity services like the one Unico provides become even more relevant in the coming years. Social networks, financial services, e-commerce, and messaging apps were all built before this AI wave, and they now require strong human accountability for the actions bots take on our behalf, as well as protection against bad actors impersonating others.

At Unico, we believe humans are the core of a safe internet, and we use advanced AI to solve this humanity dilemma of our AI era. We also believe everyone should be able to protect their systems with the same mindset. With that objective, we recently built a new solution on our platform called Face Tokenization, or IdFace: from a liveness-protected face capture, it generates a unique ID based on a person's biometric characteristics, with near-100% precision and very high recall. In the past, 1:N search and 1:1 comparison required extensive computation; Unico itself runs a very large vector database with billions of vectors. With face tokenization, identification operations that used to demand heavy computing infrastructure become a simple database lookup on the customer's side. The customer holds a token rather than a face database, so proving who someone is no longer requires storing images and computing complex biometric vectors. I foresee a world of applications that combine our liveness detection and face tokenization: validating which human stands behind an agent, preventing account farming, fighting disinformation on social networks, enabling account recovery and authentication, and even performing regular KYC. Any application that needs to authenticate users, from a dozen to hundreds of millions, can benefit from the platform we are building.

By realizing this platform, the application you are building today can securely put humans at the center of its product. Unico's small contribution to a more human centric internet.

‍