The 2026 Bund Summit, themed "Technology, Geopolitics and the Global Economy," was held in Huangpu District, Shanghai, from October 19 to 22, 2026. Yu Feng, Chief Information Officer of Guotai Haitong Securities, attended the conference and participated in a roundtable discussion.
Yu Feng stated that over the past two years, AI has achieved remarkable results in the securities industry. For application scenarios with clear rules, standardized processes and high repetition, it has already improved efficiency and generated solid returns, including the scaling of services and cost reductions. He noted that AI currently exerts influence across four areas. The first is traditional services, namely client servicing and wealth management. The second is investment research and investment banking, which were previously highly information-intensive processing tasks. The third is compliance and risk control, which are middle and back-office operations. The fourth is investment research and IT operations. From the perspective of the industry's overall landscape, apart from a small number of leading brokerages, most securities firms may still be at the stage of driving progress through selected pilot points rather than achieving large-scale application.
Looking ahead, Yu Feng believes the biggest change in AI applications will be the shift from single-point assistant-type tools to cross-functional, cross-line intelligent agent clusters. Different agents will not only be able to communicate with each other but also invoke each other's data and tools, and even learn from one another, chaining together to complete entire business processes and achieve end-to-end task collaboration.
At the same time, Yu Feng proposed that AI will not change certain fundamental functions of the financial market itself, but it will change the market's operational tempo and transmission mechanisms. Based on this understanding, future dynamics may shift from person-versus-person competition to person-versus-machine and machine-versus-machine competition. Against this backdrop, new risks may emerge, such as the homogenization of algorithms and models, the loss of control over intelligent agents' action permissions, as well as information pollution and ecosystem contamination.
Yu Feng emphasized that the financial industry fundamentally cannot afford errors and requires a zero-error financial system, so it may not be feasible to wait for the emergence of a "perfect model." Guotai Haitong's practice is to address the objectives of controlling output, delimiting actions and tracing accountability through systematic engineering methods, building a responsibility chain that spans knowledge sources, file generation, continuous correction and action functions, rather than a simple isolated technological barrier. Ultimately, it must rely on human judgment to determine direction and assume responsibility, forming a mechanism of "trustworthy content and reliable output." Currently, Guotai Haitong's AI applications have moved from single-point breakthroughs to large-scale deployment.
"Because the model itself can hardly eliminate uncertainty, we can confine AI services within a manageable, retrievable and detectable scope. At that point, the role of humans becomes decisive, and only in this way can we ensure the stable and sustained operation of the financial industry," Yu Feng added.
Comments