Central Bank Report Details AI Fraud Score for Pix

In August 2026, Banco Central do Brasil, the country's central bank, published its Pix Management Report, covering the period from 2023 to 2025, offering a review of the instant payment system's evolution and its plans for the coming years. Among the initiatives detailed is the development of a centralized, predictive fraud index built on machine learning technology, designed to raise the level of protection against scams within Pix. The information was confirmed by reporting from meutudo.com.br, published on August 14, 2026, and wproo.com, published on August 12, 2026, both citing the official central bank report.

How the system would work: machine learning and network wide pattern analysis

According to the report, the new index would combine two layers of analysis. The first uses machine learning to calculate, in real time, the probability of fraud for each transaction as it occurs. The second applies artificial intelligence to analyze historical behavior patterns across the entire Pix network, drawing on data from the Sistema de Pagamentos Instantâneos, or SPI, the instant payments system, and the Diretório de Identificadores de Contas Transacionais, or DICT, the directory of transactional account identifiers. The result would be a comprehensive risk indicator calculated for each individual transaction, factoring in not only the user's own history but the aggregate behavior of the whole network.

Real time distribution, final authorization stays with each institution

The plan calls for this risk indicator to be distributed instantly to all participating Pix institutions, including banks, fintechs, and payment institutions. Each institution could feed the index into its own fraud defense mechanisms, using it as an additional input to decide, autonomously, whether to authorize or block a specific transaction. The central bank would not take on the role of approving or rejecting payments itself: that final authority remains with each institution, which combines the new indicator with its own internal risk rules.

No confirmed timeline yet

The central bank's report does not set a launch date or a detailed timeline for rolling out the AI fraud index. The initiative is described as a project still in development, part of Pix's strategic planning for the years ahead, not as a feature already being tested or about to launch. That means the new risk scoring system is not yet in effect, and institutions and users alike should watch for future official communications from the central bank on when and how it will be implemented.

Other threads in the report: offline Pix, the Split Tributário tax mechanism, and cracking down on message field abuse

The fraud index is just one of several initiatives outlined in the report. The document also references the development of offline Pix, which would allow transactions to go through without an internet connection, and Split Tributário, a new mechanism for automatically withholding taxes on transactions. In parallel, a committee within the Fórum Pix, which brings together financial institutions and civil society representatives, has been working to curb misuse of the free text description field attached to transactions, currently exploited by scammers to send intimidation and harassment messages, often paired with transfers of just R$0.01. Together, these initiatives reflect the sheer scale Pix has reached: in 2025 alone, the system processed nearly 80 billion transactions, moving more than R$35 trillion, reaching 148 million individual users and 12.8 million active businesses.

Why this matters for Brazilian agencies and SMBs

For Brazilian businesses that receive payments via Pix, which today is nearly all of them, this announcement is one of the clearest signals yet that the country's financial infrastructure is moving toward AI native fraud defense operating at the scale of tens of billions of transactions a year. That matters especially for e-commerce operations, service businesses, and marketing agencies that handle client payments or build automated billing flows into customer journeys. A more effective, network wide risk index could mean fewer chargebacks and less exposure to the scams that currently hit small businesses hard. At the same time, it is worth thinking now about how a business's own transaction patterns might read to an automated risk model: sudden volume spikes, new payment relationships, or abrupt changes in behavior could, in principle, raise the risk score of a legitimate account. For that reason, agencies and SMBs should stay alert to communications from their banks and payment institutions as this and the report's other initiatives move forward.