Generative AI has rapidly spread in the field of brokerage investment and research in the past two years. From the first draft of research reports and conference minutes to the construction of valuation models, AI has already been deeply involved. At the same time as efficiency is improving, compliance risks also follow: individual analysts directly introduce data and opinions generated by AI into research reports without verification; furthermore, policy information and fictitious industry data created by AI are sent to the market after being signed by analysts, causing problems such as misaligned policy times and untraceable data sources in research reports. The China Securities Association recently issued the “Statistical Report on the Operation of Securities Research Reports Issued by Securities Companies in 2025”. For the first time, “Establishing an Internal Control Mechanism for Reviewing AI Investment and Research Content” is an industry standard, clearly prohibiting analysts from directly accepting unverified data, opinions, deductions, and research findings generated by AI. Although the content of this provision is simple, it is not light in weight. It redefines the position of AI and analysts in the investment and research chain: AI can be an auxiliary tool, but it cannot replace professional judgment; analysts can use technology, but they must be responsible for the final published report. The technology itself is neutral. It can amplify research efficiency as well as research errors. Only by correcting the supporting role of AI, compacting analysts' responsibility for guarding the door, and establishing a closed loop of internal control of the institution can the technical dividend be truly transformed into the quality of research.

Generative AI has rapidly spread in the field of brokerage investment and research in the past two years. From the first draft of research reports and conference minutes to the construction of valuation models, AI has already been deeply involved. At the same time as efficiency is improving, compliance risks also follow: individual analysts directly introduce data and opinions generated by AI into research reports without verification; furthermore, policy information and fictitious industry data created by AI are sent to the market after being signed by analysts, causing problems such as misaligned policy times and untraceable data sources in research reports. The China Securities Association recently issued the “Statistical Report on the Operation of Securities Research Reports Issued by Securities Companies in 2025”. For the first time, “Establishing an Internal Control Mechanism for Reviewing AI Investment and Research Content” is an industry standard, clearly prohibiting analysts from directly accepting unverified data, opinions, deductions, and research findings generated by AI. Although the content of this provision is simple, it is not light in weight. It redefines the position of AI and analysts in the investment and research chain: AI can be an auxiliary tool, but it cannot replace professional judgment; analysts can use technology, but they must be responsible for the final published report. The technology itself is neutral. It can amplify research efficiency as well as research errors. Only by correcting the supporting role of AI, compacting analysts' responsibility for guarding the door, and establishing a closed loop of internal control of the institution can the technical dividend be truly transformed into the quality of research.