Zayon's STEPA Research Selected Among Brazil's 100 National Innovation Cases 2026

STEPA, an AI architecture for early detection of structural fragility in securities issuers, was selected at the MIT-ANPEI Summit from 439 submissions.

São Paulo, Brazil, September 29, 2026: Zayon announced today that STEPA (Structural Trajectory Embedding Predictive Architecture), a research line developed by the company, has been selected among the 100 National Innovation Cases 2026 at the MIT-ANPEI Summit, out of 439 submissions.

STEPA is the core architecture of StructureWatch, a technology Zayon is developing for the early detection of structural fragility in securities issuers. The research starts from a question about how capital markets supervision works today: what if supervision could learn trajectories, instead of only looking for threshold violations?

The central idea is that an issuer is not a set of isolated indicators. It has a trajectory. Its financial structure, regulatory filings, language, disclosure patterns and relationship with the macroeconomic environment all change over time. STEPA learns that dynamic in a representation space and seeks to identify when a trajectory begins to move away from structural patterns associated with a healthy issuer.

The approach reflects three shifts in how fragility detection is framed:

• from snapshots to trajectories;

• from thresholds to structural dynamics;

• from isolated signals to contextual representation.

Technically, the architecture is based on predictive representation learning, inspired by Joint Embedding Predictive Architectures (JEPA) and world-model concepts, adapted to multidimensional time series built from financial and regulatory information.

STEPA remains in an experimental phase. Zayon has developed a proprietary architecture and an implementation is under way, investigating a clearly defined scientific hypothesis. The research has an official registered publication with a DOI, available https://crivum.org/p/xf0f7dd1

.

"STEPA started from a simple idea that is technically demanding: an issuer is not a set of isolated indicators. It has a trajectory," said Deborah Nogueira, Founder and CEO of Zayon. "This recognition puts a research line developed inside Zayon in dialogue with a much broader community of innovation, technology and markets. We are building a new way of looking at how organizations evolve, and there is still a lot to investigate."

The list of selected cases is available on the MIT-ANPEI Summit 2026 page.

About Zayon: Zayon builds Trustworthy AI Decision Infrastructure that turns AI capabilities into decisions institutions can trust, govern and audit. Its XAID architecture binds intelligence, evidence and policy into decisions that are explainable, governed and auditable, starting with financial institutions in Brazil. Zayon was founded in Brazil by specialists in economic engineering, AI and large-scale public-sector projects. Learn more at zayon.tech.

Media contact: secom@zayon.tech