Date:
Abstract: Financial market infrastructures (FMIs) – including exchanges and other trading facilities, clearing and settlement and other high-availability systems – increasingly embed AI components (algorithmic surveillance engines, AI-driven pre-trade risk systems and smart order routing) into execution-critical workflows where failures carry systemic consequences. To external observers, testing such systems presents additional challenges: they operate as black boxes, exposing no model internals, training data or decision logic. This makes the AI model-level testing methods that assume white- or grey-box access insufficient in real-world financial environments. This paper proposes a black-box testing approach grounded in passive financial protocol traffic analysis combined with property- and rules-based validation derived from exchange rulebooks, regulatory requirements and operational and protocol specifications. The passive testing approach is demonstrated against two classes of AI component behaviour: pre-trade risk decision coherence with observable position state and best-execution compliance of routing decisions across exchange venues. Results from applying the approach to protocol traffic captured in a test environment show that rules-based properties expressed over passively captured financial protocol traffic constitute a viable partial test oracle for black-box validation of financial systems with AI components, without requiring access to model internals or active test injection.
The short paper was presented at the 2026 IEEE International Conference on Artificial Intelligence Testing (AITest) that took place on 27-30 July 2026.
Keywords: black-box testing, passive testing, property-based testing, financial market infrastructure, protocol traffic analysis, AI Testing