# 监管高风险 AI 部署的关键风险阈值,及对全自动决策强制人工监督

- 日期: 2026-07-07
- 收录: 2026-07-20
- 相关方: Mr Alex Yeo · Mrs Josephine Teo
- 来源: https://sprs.parl.gov.sg/search/#/sprs3topic?reportid=written-answer-23658
- sgai: https://sgai.md/zh/debates/written-answer-23658/
- 许可：sgai 自产内容（摘要、译文、分析）CC BY 4.0，署名并链接 sgai.md 即可；逐字原文（国会记录、演讲、字幕、政策原文）© 原权利人，仅供引用。条款：https://github.com/meltflake/sgai/blob/main/DATA-LICENSE.md

## 为什么重要

杨莉明未就强制人工监督给出硬性承诺,仅以《职场公平法》等既有法兜底,AI 自动决策问责仍靠软法而非红线

## 摘要

杨陞才先生(Alex Yeo)就 NAIS 更新中"必要且有效时将监管或立法"一语,书面质询数字发展与信息部:(a) 哪些关键风险因素或阈值决定何时须对"高风险 AI 部署"进行监管;(b) 部门是否会对实质影响个人的全自动决策(如就业决定或高风险场景)强制要求"有意义的人工监督"。部长杨莉明(Josephine Teo)答复:与任何技术一样,是否监管取决于部署方式、可能造成的伤害性质,以及现有措施是否足以应对。她强调新加坡语境下许多 AI 风险已被现有法律覆盖——用 AI 招聘仍须遵守即将出台的《职场公平法》,《网络犯罪危害法》让警方可阻断 AI 诈骗;现有措施不足且能设计出合适对策时,政府会及时立法(如《选举(网络广告诚信)(修订)法》)。这些法律再辅以分行业指引,如 MAS 就金融业《AI 风险管理指引》咨询、卫生部更新《医疗 AI 指南》(AIHGle 2.0)。政府将持续研究合适的 AI 监管姿态。

## 要点

- 是否监管取决于部署方式、伤害性质与现有措施是否足够——不设一刀切的"高风险"阈值
- 许多 AI 风险已被现有法律覆盖:《职场公平法》管 AI 招聘、OCHA 阻断 AI 诈骗、选举法禁深伪
- 以分行业指引(MAS 金融 AI 风险管理、卫生部 AIHGle 2.0)补充硬法

## 全文

50 杨陞才先生（Mr Alex Yeo）就 NAIS 更新中「政府将在必要且有效时进行监管或立法」一语,询问数字发展与信息部：(a) 哪些关键风险因素或阈值决定何时须对高风险 AI 部署进行监管；(b) 部门是否会对实质影响个人的全自动决策（如就业决定或高风险场景）强制要求有意义的人工监督。

杨莉明女士（Mrs Josephine Teo）：与任何技术一样,是否需要监管,取决于其部署方式、可能造成的伤害的性质,以及现有措施是否足以应对这些伤害。

在新加坡的语境下,许多人工智能（AI）风险已通过现有立法得到应对。例如,使用 AI 工具的雇主仍须遵守即将出台的《职场公平法》中关于公平与择优雇佣的要求。同样,《网络犯罪危害法》等法律让新加坡警察部队得以预防和阻断由 AI 助长的诈骗。当现有措施被评估为不足、且能设计出合适的对策时,政府会确保其及时落地。例如,《选举（网络广告诚信）（修订）法》的通过,正是为了防止在选举期间滥用被数字生成或篡改的候选人内容。

这些法律再辅以特定行业的指引。此类指引起初未必用于强制合规,而是在监管机构考虑未来政策时提供有用反馈。例如,新加坡金融管理局已就金融服务业《人工智能风险管理指引》进行咨询,卫生部则更新了《医疗人工智能指南》（AIHGle 2.0）。

政府始终强调负责任地开发与部署 AI 的必要性。随着技术与风险格局的演变,我们将继续研究合适的 AI 监管姿态,并评估现有措施是否足够。

## Hansard (original, English)

© Parliament of Singapore — reproduced for reference only.

50 Mr Alex Yeo asked the Minister for Digital Development and Information further to the Update to NAIS, which states that Government will "regulate or legislate where necessary and effective" (a) what key risk factors or thresholds determine when a high-risk AI deployment would warrant regulation; and (b) whether the Ministry will mandate meaningful human oversight over fully automated decisions that materially affect individuals, such as employment decisions or high-risk scenarios.

Mrs Josephine Teo: As with any technology, the need for regulation depends on how it is deployed, the nature of harm that may be caused and whether existing measures are effective in addressing those harms.

In Singapore's context, many artificial intelligence (AI) risks are already addressed through existing legislation. For example, employers that use AI tools must still comply with the requirements for fair and merit-based employment practices in the upcoming Workplace Fairness Act. Similarly, laws like the Online Criminal Harms Act enable the Singapore Police Force to prevent and disrupt AI-enabled scams. Where existing measures are assessed to be inadequate and a suitable response can be designed, the Government will ensure its timely implementation. For example, the Elections (Integrity of Online Advertising) (Amendment) Act was passed to prevent the misuse of digitally generated or manipulated content of candidates during elections.

These laws are complemented by guidelines in specific sectors. Such guidelines may not be for compliance at the outset but provide regulatory agencies with useful feedback as they consider future policies. For example, the Monetary Authority of Singapore has consulted on Guidelines on AI Risk Management for the financial services sector, while the Ministry of Health has updated the AI in Healthcare Guidelines (AIHGle 2.0).

The Government has consistently emphasised the need for responsible AI development and deployment. We will continue to study the appropriate regulatory stance for AI and assess whether existing measures are adequate as the technology and risk landscape evolves.