Feedzai Fusion São Paulo puts trust at the center of agentic fraud defense

Trust will determine whether agentic finance takes hold

Leaders from financial institutions, payment networks, fintech companies, regulators and consulting firms met at 033 Rooftop on Avenida Juscelino Kubitschek on November 5. Participants included Banco do Brasil, Itaú Unibanco, Rede, Sicredi, Dock, Fiserv, Mastercard, AWS, Febraban, the Global Anti-Scam Alliance, the Brazilian Federal Police and UK Finance.

Feedzai’s Head of Global Sales, Nuno Pires, asked whether people would allow an AI agent to manage their finances, choose investments and make payments. His keynote argued that advanced technology alone does not create a new era. Adoption also depends on technology working at scale and on people being confident that problems will be resolved quickly.

Feedzai pointed to Pix, which has reached more than 170 million users, as an example of a service that solved a real problem while operating within a regulated and protected environment. In the company’s view, trust does not mean removing all risk. It means managing risk consistently at scale.

A panel featuring Leonardo Carissimi of Mastercard and Luís Pereira Lima of Itaú Unibanco and Rede examined the practical limits of agentic commerce. The discussion covered authentication, behavioral biometrics and tokenization, as well as who owns the customer relationship when agents shop on a person’s behalf.

Human judgment remains part of scam prevention

Actress, director, author and neuroscientist Danni Suzuki argued that technology’s progress makes human understanding and cooperation more important. Citing neuroscientist António Damásio’s research, she said emotions influence decisions, including decisions made by professionals such as surgeons and judges.

Her talk connected with a panel on human-centered scam prevention. Antonio Soares of Dock and Juliano Seixas of Sicredi discussed authorized push payment scams, in which the victim is the person pressing “send.” According to the event’s discussion, generic warnings can be ignored, while a timely and relevant intervention may interrupt a scam without creating unnecessary friction for legitimate customers.

Regulatory discussions led by Renata Salvini of the Global Anti-Scam Alliance and Raphael Mielli of Febraban covered the legal definition of mule accounts and emerging rules for telecoms and big tech platforms. The speakers also emphasized data sharing and continuing consumer education.

Feedzai calls for a network approach to fraud

Ruth Ray, Managing Director of Economic Crime at UK Finance, said criminals connect fraud, money laundering, sanctions evasion, cybercrime and human trafficking even when institutions and laws treat them separately. She described how a scam targeting someone in Brazil can involve scam compounds, mule accounts, shell companies and cryptocurrency.

Ray also discussed the UK’s mandatory reimbursement regime for authorized push payment scams, which has been in force since October 2024. She said victims receive more consistent outcomes and their money back, but argued that the regime has not stopped fraud because money still reaches criminals.

Her priorities were to build risk controls into products and payment journeys, connect intelligence so organizations can act before funds disappear, and measure disruption rather than activity. Other panels explored breaking down channel silos, creating a real-time customer view and monitoring incoming payments as well as outgoing ones.

Feedzai outlines its agentic RiskOps strategy

Feedzai Co-founder and Chief Science Officer Pedro Bizarro said AI agents will change the economics of fraud management because consumers may allow them to log in, pay bills and move money. Fraudsters, he said, will use similar agents.

Feedzai described three roles for itself: protecting against agentic attacks by detecting agents and classifying their intent, enabling banks to connect their own or partner-built agents through Bring Your Own Agent, and providing agents inside the tools used by fraud and anti-money laundering teams.

The company said its first such agent, Farol, is already live in RiskOps Studio. Analysts can communicate with it in plain language, and it selects skills for tasks such as identifying rules that have not caught fraud in 30 days and drafting a suspicious activity report.

For users and financial institutions, the event’s message is that AI-enabled finance will require more than capable models. It will depend on trusted controls, shared intelligence, timely human-centered interventions and a view of fraud that spans the wider network.