Guojin Securities CIO Wang Hongtao: AI will reshape the securities industry on three levels

Ask AI · How does Guotai Junan Securities’ AI practice balance innovation and risk?

China Financial Times (Cailianshe) March 27 News (Reporter Wang Yuling) Currently, AI intelligent tools have penetrated every industry at an unprecedented pace. Behind the gains in efficiency lies a widespread anxiety among practitioners—who will be replaced? Who will be able to stay? Faced with this unavoidable transformation, Wang Hongtao, Chief Information Officer of Guotai Junan Securities, has offered his answer:

“How you use AI, AI helps you achieve. You give it meaning; it lights up your road ahead.”

From a “hand each OpenClaw” to an industry-wide ban, what Wang Hongtao sees is the industry’s common consensus: it’s not a question of whether to use AI, but how to use it safely and in a controllable manner.

For the securities industry, AI’s impact will undergo three levels of leapfrogging: operational enhancement (Do Better), business innovation (Do New), and ecosystem reshaping (New Game). Ultimately, AI will evolve from a “back-office tool” into an “AI digital coworker” working side by side with employees.

OpenClaw Turmoil: The “ban” is not shortsighted caution, but fastening the seatbelt while running at full speed

Since the Spring Festival, powered by the “brain-hand integration” capability of its intelligent agents, OpenClaw has quickly become the open-source software with the most stars on GitHub, even surpassing Linux—an open-source project with decades of history. Many practitioners in the securities industry, at first glance, “raised one shrimp each,” excited by AI’s leap from “conversation” to “execution.” However, just a few days later, safety concerns surfaced, and the industry swiftly issued risk alerts again, shifting OpenClaw toward an “outright ban.”

Wang Hongtao said that while others are seeing “a ban,” he is seeing the parallel of “prohibition” and “exploration”—on one side, multiple brokerages issued internal notices forbidding private installation; on the other, at least 8 brokerages, including Guotai Junan, have issued intensive special research reports to deeply interpret OpenClaw’s industrial value.”

Wang Hongtao analyzed that this seemingly contradictory phenomenon reflects the industry’s prevailing mindset in exploring AI applications: on one hand, enterprises and employees have a strong drive to improve their own competitiveness and efficiency; on the other, enterprises and regulators are deeply concerned about threats such as systems going out of control, information security, and sensitive data leakage. However, this round of “bans” is not a comprehensive prohibition on use—the “ban” is on using OpenClaw in environments that may involve sensitive information and may trigger security incidents without strict security assessment. For exploration and use within sandbox and controllable environments, the industry actually adopts an attitude of allowing, and even encouraging it.

So, Wang Hongtao believes this round of “ban” is not shortsighted caution, but fastening the seatbelt while running at full speed. A common industry understanding is forming: it’s not a question of whether to use AI, but how to use it safely and in a controllable manner.

Four major technical risks of OpenClaw in financial scenarios

OpenClaw’s tendency to trigger security concerns mainly comes from its features of “local high privileges” and “autonomous operations.” Wang Hongtao distilled the risk essence in a single line: “When AI turns from a ‘staff assistant’ into an ‘executioner with the key,’ the cost of losing control will be exponentially amplified.”

In specific financial scenarios, he summarized four major technical risks:

First is unclear permission boundaries. To achieve strong execution power, OpenClaw is default granted extremely high local system privileges; once it is attacked or induced, it can perform actions beyond its authorization.

Second is the risk of command injection. Attackers can induce the AI to read and execute over-privileged instructions by crafting malicious webpages or files.

Third is the plugin ecosystem going out of control. Some third-party Skill packages have been confirmed to contain malicious code; after installation they can steal keys, and after deployment they can plant backdoors.

Fourth is mis-operations caused by AI hallucinations. Outputs from large models have randomness; if instructions are executed directly without review, they may cause irreversible losses.

“Technology itself is neutral, but when deploying it in financial scenarios, ‘controllability’ must be a prerequisite for ‘usability,’” Wang Hongtao said. “During innovation, only by ensuring security and stability can we move steadily and go far.”

OpenClaw deployment: from “ban” to “governance control” to “reasonable use”

For the future applications of tools like OpenClaw, Wang Hongtao’s judgment is that the industry will experience an evolution from “prohibition to governance control, from governance control to reasonable use.”

“OpenClaw is not unusable; the key is incorporating it into official system workflows and pairing it with well-developed risk prevention and control measures,” he emphasized. In future OpenClaw deployments, the core is three keywords: clear permission boundaries, robust audit mechanisms, and secure isolation.

From a practical application perspective, the deployment value of OpenClaw is mainly reflected in three directions:

First is information acquisition. Free human staff from repetitive information collection, sorting, and monitoring work—so that AI becomes a 24-hour online “intelligence assistant.”

Second is system operations. Embed AI into operation and maintenance, office, and other system operation workflows to enable command-style interactions and automated execution, making operations more efficient and more convenient.

Third is knowledge management. Activate knowledge assets such as research reports, documents, and institutional rules accumulated within organizations; use natural-language interaction to precisely retrieve and integrate them, turning knowledge from “sleeping” into “live water.”

These directions point to one underlying essence: to let AI take on repetitive cognitive work that humans are not good at or are unwilling to do, freeing people to do things with true creative value.

He revealed that Guotai Junan Securities plans to first pilot in office assistance and non-sensitive scenarios to validate the results before extending to middle and back-office areas. For high-risk scenarios such as trade execution, it will remain cautious—only after undergoing a strict “manual review” mechanism can it be allowed.

“Future industry scenarios won’t be ‘one wild shrimp per person,’ but rather standardized applications of ‘graded feeding and controllable release,’” Wang Hongtao said.

Guotai Junan Securities’ AI path: from “pilot exploration” to “AI digital employees”

The popularity of OpenClaw has validated Guotai Junan Securities’ assessment made as early as May 2025—that large models could be a new generation of operating systems.

“Within this space, Guotai Junan Securities is not a bystander, but an active explorer and practitioner,” Wang Hongtao introduced. As early as 2024, Guotai Junan Securities had already achieved preliminary exploration of job intelligence: automatic generation of industry-chain knowledge graphs, and automated data cleansing, compressing researchers’ data preparation time from 2-3 weeks to 1-2 days.

In 2025, Guotai Junan further carried out some distinctive explorations—for example, building a “large-model quantitative trading system.” It lets multiple large models compete side by side, processing information, forming decisions, and executing them in real trading environments, thereby enabling trade transparency and cross-model comparison. It also built systems such as a “large-model valuation framework for listed companies.”

By 2026, Guotai Junan Securities is comprehensively advancing a more systematic initiative—job intelligence transformation actions, namely the systematic construction of Guotai Junan AI digital employees.

“Our philosophy is: AI is not a replacement for people, but an amplifier of people’s capabilities,” Wang Hongtao said. With full security assessments as a prerequisite, this year has already launched job-intelligence transformation covering multiple scenarios such as human resources, risk control, investment banking, proprietary trading, and investments. It aims to build a “self-evolving business agent group,” allowing employees and agents to grow in sync.

When discussing Guotai Junan’s achievements in AI applications, Wang Hongtao summarized them across four dimensions:

First, on the customer service layer, the AI investment advisory system has formed a large-scale effect.

In May 2025, Guotai Junan Securities took the lead in releasing an AI investment advisory product. It now covers five core products: AI stock selection, AI holding optimization, AI stock unlocking, AI fund selection, and AI fund holding optimization. After 43 iterations throughout the year, the number of registered users has exceeded 600k; the number of signed and retained account holders is over 340k; the signed equity asset under custody amounts to 31.5 billion; and value-added revenue exceeds 30 million. At the same time, intelligent agents provide customers with intelligent companion services through three layers of identity: “AI customer service, AI advisors, and AI assistants.” Currently, the daily active usage of agents has reached more than 15k. Compared with the original AI customer service, it is 10 times higher. The recognition rate for customer-service questions rose from 50% to 98%, and the rate of escalation to humans dropped by 4 times.

Second, on the business enablement layer, large models have been fully embedded into all business lines across the company.

Currently, large models have been deployed in more than 60 scenarios, including wealth management, institutional clients, proprietary trading, investment banking, functional lines, and subsidiaries. In the investment banking domain, an intelligent agent group based on large models has been built to automatically generate due diligence reports and intelligent valuations. In connection with valuations of listed companies, multiple national invention patents have been applied for; valuation accuracy and explainability are both superior to traditional approaches. In the proprietary trading domain, 25 AI investment research and office automation tools have been deployed; cumulative usage is nearly 20k times, saving labor costs equivalent to more than 40k working hours. Scenarios such as fixed-income quarterly report analysis have significantly released research and trading capacity. In the institutional domain, an assistant for industrial-chain risk transmission analysis has been built, which can intelligently identify the upstream and downstream networks and simulate risk transmission paths. In the wealth management domain, an intelligent risk command center has been built, adding more than 100 core risk factors; it enables real-time monitoring and triggering of multi-dimensional public-opinion risk, reinforcing the defense line of product quality for value-added offerings.

Third, on the organizational efficiency layer, large models have become an “AI coworker” for all-employee office work.

The company has launched tools such as Guotai Junan Qianwen, GPT employee assistants, intelligent document recognition, and Xiaojin desktop assistants. The cumulative employee usage rate exceeds 95%. On the technology lines, through AI programming assistants and an AI DevOps delivery platform, overall coding efficiency has increased by about 40%, trouble-shooting and repair time has decreased by 70%, and UI design efficiency has increased by 30%-50%.

Fourth, in original research and scientific inquiry, it continuously accumulates industry influence.

Guotai Junan Securities received the 2023 Financial Technology Development Award. It also holds the industry’s first invention patent for “securities + large models.” At the same time, it continues to produce original outcomes such as papers, patents, research topics, and standards, and it hosts the financial AI sub-forum of the Technology Conference of the Greater Bay Area Exchanges.

“Guotai Junan Securities’ AI application achievements are not just stacking concepts; they are real-world deployment practices embedded into business, serving customers, and enabling employees,” Wang Hongtao said. “We are moving from single-point innovation toward systematic accumulation, and from tool enablement toward a leap in value.”

AI reshapes the securities industry: three levels of leapfrogging

At a big-picture level, Wang Hongtao believes AI’s impact on the securities industry will go through three levels of leapfrogging:

Operational enhancement (Do Better): use AI to improve efficiency and reduce costs without changing the essence of the business.

Business innovation (Do New): to create entirely new services that were previously impossible to offer.

Ecosystem reshaping (New Game): to rebuild the industry’s value chain.

Ultimately, AI will evolve from a “back-office tool” into an “AI digital coworker” working alongside employees.

In terms of each business line, he outlines a clear picture of transformation:

Proprietary trading and investment lines: AI will move from “assisting analysis” to “assisting decision-making.” The large-model quantitative trading system Guotai Junan has built enables multiple models to compete side by side, handling information and generating signals in real market environments.

Investment banking: AI will reconstruct due diligence and filing processes. From drafting prospectuses to dynamically tracking new regulatory rules, to intelligent organization of due diligence materials—investment banking staff can focus on deal structure and client communication.

Investment advisory: AI will deliver true “one-size-doesn’t-fit-all, tailored to each person.” Guotai Junan’s AI investment advisory has served tens of thousands of users, bringing strategy capabilities that were originally for institutions down to individuals.

Institutional services: AI will enhance the depth of services for professional institutions. By generating intelligent research reports, conducting customized data mining, and preparing materials for roadshows automatically, institutional clients can enjoy more efficient and more precise research services.

Human resources and training line: AI will become a “super trainer.” Through AI study communities and AI academies, it enables personalized learning path recommendations and intelligent Q&A mentoring, so every new employee has an AI mentor accompanying them throughout.

Compliance and risk control line: AI will move from “after-the-fact traceability” to “real-time early warning.” Through AI Agents performing targeted retrieval of new regulatory rules, automatic monitoring of abnormal trades, and dynamic detection of liquidity risk, compliance can get ahead of business—from “putting out fires” to “preventing fires.”

IT line: AI will reshape development and operations modes. From “Vibe Coding” to dramatically boost development efficiency, to AI Agents automatically巡检 servers, diagnose faults, and autonomously repair—every IT professional will become a “full-stack engineer.”

“ The biggest change brought by AI is not replacing anyone, but enabling people in every business line to be freed from repetitive work, and to give time back to truly value-creating activities,” Wang Hongtao said.

Under AI anxiety, how should practitioners position themselves?

The shockwave of AI is moving too fast, bringing widespread industry anxiety. In this era of humans dancing with AI, how should we place ourselves?

“AI anxiety—I fully understand and empathize,” Wang Hongtao admitted. “Every time there is a technological revolution, some people are replaced. More people worry about being replaced. But this revolution is faster, broader, and more impactful.”

He shared several suggestions:

First, AI is an amplifier of experience and capability. Continuously improving your own professional and cognitive abilities is always extremely important.

Second, get deeply involved: use AI, understand AI, and recognize AI.

Third, understand what changes and what stays the same in the AI era. The value of large-scale repetitive work that some AIs are good at will weaken, while trust, responsibility, and emotional connections in a human-led society will remain valuable.

Fourth, adopt an open mindset, embrace change, and build resilience.

“AI is not just a tool—it’s an extension of your thinking; it’s the projection of your own creativity into the digital world. The value of AI is not in itself, but in how you use it.” In closing, he summarized the best for this interview with one final line:

“How you use AI, AI helps you achieve. You give it meaning; it lights up your road ahead.”

(Cailianshe reporter Wang Yuling)