资本市场中的AI：在创新与诚信间寻求平衡

In shortIn this exploratory study, AFM presents how AI is reshaping every stage of the trading lifecycle: pre-trade, execution, and post-trade, bringing opportunities for insight and efficiency. At the same time, it amplifies existing market integrity risks while creating newones. Recognising that further research is needed in this fast-moving domain, market functioning will rely even more on the design of Executive summary Well-functioning capital markets are essential to the real economy,supporting household wellbeing and broader economic health. Three conclusions stand out:1.Market integrity is shaped by human choices embedded in AI systems, especially via model objectives and constraints, thecontext in which they operate, and the data they ingest.As modelsbecome more autonomous, outcomes increasingly depend on themodel design, the system in which they operate, the data quality, When they are efficient and robust, they enable wealth formation,facilitate risk-sharing, and allocate capital to its most productive use.AI is increasingly integral to capital markets and is expected to remaina defining element of modern market functioning. Used responsiblyand governed with clear principles, AI can strengthen price formation, Yet the same capabilities that make AI transformative also make itvulnerable.Adaptive models learn quickly and at scale, stretchingtraditional oversight. They depend on models' incentives whoseintegrity is not always assured, and may optimise in ways that arenarrow, opaque, or unpredictable. Even when individual actors followsound practices, their models interact within a tightly coupled system; 2.Risks emerge not only from the individual model but also fromthe environment in which models operate.System interdependenciescan amplify impact as shared inputs and optimisation targets may 3.Market participants remain fully accountable for the outcomesof their systems, regardless of technological complexity.The useof self-learning models does not dilute this responsibility, marketparticipants remain answerable for how their models behave. For market participants, AI reshapes not only decisions, but alsohow they can be justified, audited, and governed.For retail investors,general purpose AI tools may appear authoritative without theprotections of regulated advice.For supervisors, the shift to self-learning models challenges assumptions about explainability and The future of AI-driven capital marketsmay evolve into a mixedecosystem in which trusted, well-governed AI models interact withless trusted and opaque ones, while capabilities and use-cases develop Achieving this will require human oversight, transparency, andaccountability to evolve accordingly so that AI can realise its The AFM’s aim is to remain agile and innovation-aware: promotingtrustworthy AI as the norm while monitoring and mitigatingvulnerabilities from ungoverned autonomy, unstable interactions, Based on these findings, the AFM recognises the following priorities 1.Trusted AI models as a foundation of market integrity:theAFM aims for markets where AI models are reliable and safe bydesign. Sound model validation, clear safeguards and strongdata governance will shape competitive dynamics, as trust in AIbehaviour becomes a meaningful differentiator. Accordingly, the 2.Supervising a mixed ecosystem of trusted and untrusted AI systems:some participants will operate highly governed and transparentAI systems, while others may rely on opaque or unstable models.The AFM aims to adapt supervision accordingly: well-governedAI can be met with more predictable, proportionate expectations,while high-risk or opaque systems require closer scrutiny, creatingthe right incentive structure. Moreover, it is important to reassesswhether today’s capital market infrastructure is still calibrated for 3.Addressing system-level dynamics and potential feedbackloops:market dynamics increasingly depend on how AI models behave both individually and in combination. Understanding theinteractions between them is essential to identifying conditions thatcan generate self-reinforcing feedback loops and amplify marketstress. In this context, a broader debate is needed. The question is Executive summary2 Introduction6 1.Making or breaking markets8 3.Pre-trade 3.1Pre-trade opportunities.....................................................................163.2Pre-trade risks......................................................................................17 4.Execution 4.1Execution opportunities..................................................................194.2Execution risks....................................................................................20 Conclusions Introduction The ambition to create autonomous systems is not new. In Greekmythology, Talos, the bronze giant forged by Hephaestus to patrolCrete, embodied both the promise and the peril of artificial autonomy.Designed to protect the island without human intervention, Ta