中文导读

新加坡金融管理局10月7日发布金融业AI风险管理指引。本文通过低风险工具、关键系统和第三方服务的比较，解释为什么控制措施应随实际风险变化。

正文

Imagine that a bank uses artificial intelligence to write a short office note. Now imagine another AI system that can take action and affect a customer’s money. Should both systems face exactly the same checks? New rules from Singapore say no. The level of control should match the possible harm.

本段译文

设想一家银行用人工智能撰写一份简短的办公室便笺。再设想另一个AI系统能够采取行动并影响客户的资金。两种系统是否应该面对完全相同的检查？新加坡的新规则认为不应该。管控程度应与可能造成的损害相匹配。

On October 7, the Monetary Authority of Singapore, or MAS, published guidelines for AI risk management in financial institutions. They apply to every kind of financial institution and every form of AI. However, the rules follow a risk-based approach. A simple tool may need basic policies if its failure would have little effect. A more important system needs stronger controls.

本段译文

10月7日，新加坡金融管理局发布了面向金融机构的AI风险管理指引。指引适用于每一类金融机构和每一种AI。不过，规则采用基于风险的方法。如果一种简单工具失效后影响很小，基本政策可能已经足够；更重要的系统则需要更严格的控制。

The first task is knowing where AI is being used. Institutions should keep a suitable list of their AI systems and judge the risk of each case. They should check data, testing, human oversight, cybersecurity, monitoring and changes during the system’s full life. Boards and senior managers must also set clear duties. MAS says a special AI committee is not always necessary if current management structures can do the job.

本段译文

首要任务是了解AI被用在什么地方。机构应保留一份适当的AI系统清单，并判断每个使用场景的风险。在系统的整个生命周期内，它们应检查数据、测试、人工监督、网络安全、监测及系统变更。董事会和高级管理人员也必须明确职责。新加坡金融管理局表示，如果现有管理结构能够完成工作，并不一定需要专门成立AI委员会。

Responsibility does not disappear when an outside company supplies the technology. A bank or insurer must still decide whether the third-party AI is fit for its purpose. It must seek enough information from the supplier and add other controls when there are gaps. If the risk remains too high, the institution should limit, stop or replace the service.

本段译文

采用外部公司提供的技术，并不会让责任消失。银行或保险公司仍必须判断第三方AI是否适合预定用途。机构需要从供应商获得足够信息，并在存在缺口时增加其他控制。如果风险仍然过高，机构应限制、停止或更换该服务。

Autonomous AI makes this question more urgent. Such systems can carry out actions, work with outside tools and change their behaviour with less human oversight. CNA reported that Singapore’s government is running trials to learn how to use these systems responsibly. MAS also plans to ask the financial industry in 2027 what extra guidance on autonomous AI would be useful.

本段译文

自主AI让这个问题变得更加紧迫。这类系统能够执行操作、使用外部工具，并在较少人工监督的情况下改变行为。亚洲新闻台报道，新加坡政府正在进行试验，以了解如何负责任地使用这些系统。新加坡金融管理局还计划在2027年向金融业征询意见，了解自主AI还需要哪些补充指导。

The guidelines were published now, but they do not take effect immediately. The first parts will apply from October 7, 2027. Later sections must be followed by October 7, 2028. This phased timetable gives institutions time to find their AI systems, assign responsibility and improve their controls.

本段译文

指引现在已经发布，但不会立即生效。第一部分将从2027年10月7日起实施，后续部分须在2028年10月7日前落实。分阶段的时间表让机构有时间找出自己使用的AI系统、分配责任并改进控制措施。

The main idea is not that every use of AI is equally dangerous. It is that the same technology can create very different results in different jobs. By asking institutions to match controls to real risk, Singapore is trying to protect customers without blocking useful tools. Whether the approach works will depend on careful action, regular review and clear evidence after the rules begin.

本段译文

核心观点并不是所有AI用途都同样危险，而是相同技术在不同工作中可能产生很不一样的结果。新加坡要求机构让控制措施与实际风险相匹配，希望在不阻碍实用工具的同时保护客户。这种方法是否有效，将取决于规则实施后的谨慎行动、定期审查和清楚证据。

新闻来源

• Monetary Authority of Singapore · MAS Sets Out Supervisory Expectations on Responsible AI Adoption by Financial Institutions
• CNA · Singapore government running trials to understand how to implement AI agents responsibly: Tan Kiat How
• 联合早报 · 金管局发布AI风险管理指引 机构用第三方AI须担主责

✳更多文章 →