arXiv: FIDES: A Concordance Protocol for LLM-Generated Trading Strategies

The publication introduces FIDES, a technical concordance protocol designed to verify that AI-generated trading strategies align with the stated risk parameters and regulatory constraints of the deploying firm. It is not a new law or regulation, but a proposed technical standard for auditing the decision-making logic of large language models used in financial markets. The protocol aims to create a verifiable chain between the model's output, the underlying data, and the compliance rules it was supposed to follow, addressing the current "black box" problem in algorithmic trading.

This change directly affects financial institutions, including banks, hedge funds, and asset managers, that are using or planning to use LLMs for trade execution, portfolio rebalancing, or strategy generation. It also impacts technology vendors supplying these models and any compliance officers responsible for market conduct under MiFID II, MAR, or the EU AI Act. The protocol is particularly relevant for firms that must demonstrate explainability and auditability of automated decisions to regulators.

Compliance teams should treat this as a forward-looking signal. They should begin by inventorying any LLM-based trading tools in their environment and mapping them against existing algorithmic trading governance requirements. Next, they should evaluate whether their current testing and validation frameworks can accommodate a concordance check like FIDES, and consider piloting it in a sandbox environment. Finally, they should monitor the academic and regulatory uptake of this protocol, as it may inform future technical standards for AI audits under the EU AI Act's high-risk classification for financial systems.