
周一,一场软件故障扰乱了美国多家大型银行的服务,同时美联储一位理事警告称,日益自主的人工智能系统会带来风险。在此背景下,监管机构和行业领袖齐聚香港,讨论如何最好地监管这项技术在全球金融领域的迅速蔓延。
As a software glitch disrupted services at several major US banks on Monday and a US Federal Reserve governor warned about the risks posed by increasingly autonomous artificial intelligence systems, regulators and industry leaders gathered in Hong Kong to debate how best to oversee the technology’s rapid spread through global finance.
周一,在香港举行的彭博全球监管论坛上,包括香港证券及期货事务监察委员会行政总裁梁凤仪在内的全球顶级监管者讨论了从市场波动到网络安全风险等一系列挑战。
Top regulators from around the world, including Julia Leung Fung-yee, CEO of Hong Kong’s Securities and Futures Commission, discussed challenges ranging from market volatility to cybersecurity risks at the Bloomberg Global Regulatory Forum on Monday.
渣打银行融资与证券服务数据及数字主管Allan Song表示:“同行之间以及与监管机构之间的这种合作……不能变成一场比拼谁是最佳欺诈检测者的竞赛。”他呼吁加强跨行业和国际合作。
“This collaboration across our peers and with regulators … cannot be a competition about who is the best fraud detector,” said Allan Song, head of data and digital for financing and securities services at Standard Chartered Bank, as he called for greater cross-sectoral and international cooperation.
“这有点像一场军备竞赛,”他说,并警告称“不法分子”可能获得相同甚至更先进的人工智能模型,却不受类似的监管约束。
“It is a bit of an arms race,” he said, warning that “bad actors” could gain access to the same AI models, or even more advanced ones, without facing similar regulatory constraints.
Song表示,代理式人工智能——能够在有限人工输入下追求目标并执行多步骤任务——代表着金融业的“下一个前沿”。他补充说,该行业的角色正朝着“提供受监管的市场基础设施”演变,供人工智能代理在其中运作。
Song said agentic AI – which could pursue goals and perform multistep tasks with limited human input – represented the “next horizon” for the financial industry. The sector’s role, he added, was evolving towards “providing the regulated market infrastructure” on which AI agents could operate.
但财政司司长陈茂波强调,金融机构对其人工智能系统的行为仍负有最终责任。他补充说,人工智能代理“没有牌照可失去,也没有声誉需要保护”。
But Financial Secretary Paul Chan Mo-po stressed that financial institutions remained ultimately responsible for the actions of their AI systems. AI agents had “no licence to lose and no reputation to protect”, he added.
陈茂波发表上述言论之际,美联储理事迈克尔·巴尔在米兰的一次会议上警告称,代理式人工智能可能在压力时期加速存款外流,从而放大金融不稳定,并可能使传统流动性管理工具和监管响应时间不堪重负。
Chan’s remarks came as Fed governor Michael Barr, speaking at a conference in Milan, warned that agentic AI could amplify financial instability by accelerating deposit outflows during periods of stress, potentially overwhelming traditional liquidity management tools and regulatory response timelines.
在代理式人工智能的背景下,监管必须随着技术发展而演进,这一点很重要Giorgio Valente,香港货币及金融研究院 周一,技术故障影响了美国多家大型银行的数千名客户,包括摩根大通、美国银行和富国银行,这些银行均报告客户在访问网上银行服务时遇到问题。
In the context of agentic AI, it is important that the oversight has to evolve with the technologyGiorgio Valente, HKIMR On Monday, technology disruptions affected thousands of customers at major US banks – including JPMorgan Chase, Bank of America and Wells Fargo – who reported issues accessing online banking services.
尽管摩根大通澄清说,问题是由一个临时软件缺陷而非网络攻击造成的,但这一事件凸显了对该行业技术转型加强监管的紧迫性。
Although JPMorgan clarified that a temporary software bug, rather than a cyberattack, had caused the issue, the incident underlined the urgency for better oversight of the technological transformation of the industry.
在香港,生成式人工智能的使用已日益普及。根据香港货币及金融研究院(HKIMR)周二发布的一份研究报告,90%的受访机构表示已采用或计划采用这项技术。
In Hong Kong, the use of generative AI is already becoming widespread. According to a research report released on Tuesday by the Hong Kong Institute for Monetary and Financial Research (HKIMR), 90 per cent of surveyed institutions said they had adopted or were planning to adopt the technology.
该报告调查了香港47家金融机构,并纳入了对行业参与者19次访谈的见解。对许多公司而言,下一步是采用代理式人工智能。
The report surveyed 47 financial institutions in the city and incorporated insights from 19 interviews with industry participants. For many firms, the next step is the adoption of agentic AI.
“在代理式人工智能的背景下,监管必须随着技术发展而演进,这一点很重要,”香港货币及金融研究院主管Giorgio Valente说。他强调“人类指挥”方式的重要性,即由人类对人工智能代理的自主程度设定限制和防护栏。
“In the context of agentic AI, it is important that the oversight has to evolve with the technology,” said Giorgio Valente, head of HKIMR. He stressed the importance of a “human-in-command” approach, under which people set limits and guard rails on the extent of an AI agent’s autonomy.
根据该报告,尽管只有13%的受访者表示已采用代理式人工智能,但另有56%的受访者表示计划实施。
Although only 13 per cent of respondents said they had already adopted agentic AI, an additional 56 per cent said they planned to implement it, according to the report.
然而,采用情况仍不均衡。在规模较小的公司中,50%表示没有立即采用的计划。
Adoption remains uneven, however. Among smaller firms, 50 per cent said they had no immediate plans for adoption.