MUFG: Bank of England's Quantitative Tightening Reform Sets High Standards, QT Shifts to 'Back-Office Automation'

Lee Hardman, a strategist at Mitsubishi UFJ Financial Group, stated that the Bank of England's reform of its quantitative tightening (QT) framework marks a shift from annual reviews to a pre-announced multi-year exit plan, setting a 'very high bar' for future adjustments to the exit path. Hardman noted that the exit mechanism would only be reconsidered if the bank rate proved insufficient to achieve inflation targets or if the UK gilt market faced severe stress. He believes the reform makes QT a gradual, predictable, and largely automated process, effectively treating it as a 'back-office balance sheet operation' rather than an active policy tool. Compared to past annual decisions focused on the pace of balance sheet reduction, the new framework reflects a stronger commitment and places greater emphasis on the final target state of the balance sheet. (Sina Finance)

Four Major AI Programming Agents Expose Same Security Vulnerability, Risk of 'Skills' Update Mechanism Hijacking

Cybersecurity startup Air exclusively revealed that four mainstream AI programming agents—Anthropic Claude Code, OpenAI Codex, Google Gemini CLI, and GitHub Copilot—share the same logical flaw: their automatic 'Skills' update mechanisms fail to verify content changes, allowing attackers to bypass scans by posing as malicious updates with identical names and silently steal corporate code or intellectual property. The vulnerability stems from identical verification mechanism designs across vendors, highlighting a common industry blind spot. Currently, all but GitHub have patched the issue; GitHub claims its platform mechanisms can block such attacks but has not confirmed patch status. (Sina Finance)

Sugon and China-ASEAN AI Application Cooperation Center Sign Strategic Cooperation Agreement

On September 17, during the 23rd China-ASEAN Expo, Sugon and the China-ASEAN AI Application Cooperation Center signed a strategic cooperation agreement in Nanning, Guangxi. The two parties will advance collaborative computing-electricity integration and computing-network convergence, explore the construction of large-scale domestically produced intelligent computing clusters and their integration into the national unified computing power network, coordinate computing power and data center construction and operations across Guangxi and ASEAN countries, while expanding cross-border channels and international data hubs and other infrastructure around cross-border computing services. Sugon's subsidiary, Sugon Cooling, has already implemented multiple liquid cooling projects in ASEAN countries, including a large-scale AI intelligent computing liquid cooling project in Malaysia, and formed a regional service system through its Singapore headquarters, Malaysian laboratory, and Qingdao factory. (36Kr)

NEEQ IPO Pipeline Expands Further, 223 Listed Companies Enter Innovative Layer This Year

The NEEQ IPO pipeline continues to grow, with 12 newly listed companies on the National Equities Exchange and Quotations (NEEQ) entering the innovative layer. According to an announcement by the National Equities Exchange and Quotations, on September 17, 12 NEEQ-listed companies, including Gudeng Machinery and Yunshang Technology, were transferred to the innovative layer. As of that day, the NEEQ had a total of 5,786 listed enterprises, comprising 3,501 in the base layer and 2,285 in the innovative layer. Under the schedule, the NEEQ conducts six layer adjustments annually, with this marking the fourth adjustment for the year. Including this batch, 223 listed companies have been transferred to the innovative layer this year. According to NEEQ listing rules, issuers must be innovative layer enterprises that have been continuously listed for 12 months. (Yicai)

Overseas Highlights

OpenAI Launches Legal AI Platform Astra for Law

On September 17 (local time), OpenAI announced the launch of Astra for Law, an AI platform tailored for the legal industry. Built on OpenAI's latest model, GPT-6 Astra, the platform is designed specifically for law firms and legal tech companies. (Sina Finance)

Meta-Linked Data Center Enters U.S. Junk Bond Market for First Time

CleanSpark Inc. is seeking to raise approximately $2.23 billion through its inaugural junk bond issuance to fund the construction of AI infrastructure linked to Meta Platforms Inc. According to sources familiar with the matter, preliminary pricing discussions for the five-year notes suggest a yield of around 8.5%, roughly 2 percentage points higher than the average yield on BB-rated corporate bonds tracked by Bloomberg, reflecting investor demands for a higher risk premium on data center financing. By Thursday, when the transaction launched, investor orders had already matched the issuance size. According to compiled data, this marks the first junk bond issuance linked to a Meta data center. (Sina Finance)

Vertical Data and NVIDIA Announce CAD 189 Million AI Infrastructure Deployment in Canada

Vertical Data and NVIDIA announced the deployment of CAD 189 million worth of AI infrastructure in Canada, utilizing GB300 products. (Sina Finance)