Skip to content

BEEAI · RESPONSIBLE AI

违规与监管Industry Regulation

读懂 AI 监管,让创新有边界,让应用有底气。Understand AI regulation. Build responsibly and deploy with confidence.

新加坡专题 · 核对日期:2026年11月18日Singapore edition · Checked: 18 November 2026
本页为学习性摘要,并非完整法规清单或法律意见。适用义务取决于行业、机构身份、数据及使用场景;实际项目请核对官方原文并咨询合规或法律专业人士。This learning summary is not an exhaustive register or legal advice. Duties depend on sector, entity, data and use case. Check official texts and seek qualified advice for your project.
近期案例RECENT CASES

新加坡近期数据泄露与 AI 安全事件Recent Singapore Data Breaches & AI Security Incidents

公开信息摘要 · 重点关注 AI 使用、第三方平台、云环境及软件供应链风险。Public-source summaries covering AI use, third-party platforms, cloud environments and software supply chains.

AI 使用风险AI USE

美珍香:95,364 个电邮地址暴露Bee Cheng Hiang: 95,364 emails exposed

AI 群发程序未经充分检查,收件人电邮地址互相可见。An insufficiently checked AI email tool exposed recipients' addresses.

启示:Lesson: AI 代码先测试、再审批。Test and approve AI code.

查看 CNA 报道Read CNA report →
电信数据TELCO DATA

M1:客户身份与联系资料泄露M1: customer identity and contact data exposed

姓名、证件号码、电话、电邮及住址等资料受影响。Identity and contact details, including ID numbers, were affected.

启示:Lesson: 少存数据,严控权限。Minimise data and restrict access.

查看 M1 公告Read M1 notice →
第三方平台THIRD PARTY

新航:约 46,000 条订位记录被访问SIA: about 46,000 booking records accessed

订位供应商平台遭入侵,乘客姓名、行程等资料受影响。A booking vendor breach exposed passenger and itinerary data.

启示:Lesson: 供应商也要纳入风控。Include vendors in risk controls.

查看 CNA 报道Read CNA report →
云端测试数据CLOUD TEST DATA

SLA / IBM:约 70,000 人资料受影响SLA / IBM: about 70,000 people affected

测试云环境遭访问,未完全匿名化的个人资料外泄。A test cloud exposed personal data that was not fully anonymised.

启示:Lesson: 测试数据必须真正匿名。Fully anonymise test data.

查看 SLA 公告Read SLA notice →
AI 供应链AI SUPPLY CHAIN

LiteLLM:开源 AI 软件更新遭植入恶意代码LiteLLM: malicious code inserted into AI software updates

泄露的登录资料被用于在软件更新中植入恶意代码。Stolen credentials were used to insert malicious code into an update.

启示:Lesson: 锁定版本,扫描依赖。Pin versions and scan dependencies.

查看 MDDI 回应Read MDDI response →
重点先看KEY CONSEQUENCES

违规后果Consequences of non-compliance

先分清适用法律,再看处罚上限;不是所有AI问题都适用同一种处罚。Identify the applicable law first. Not every AI issue attracts the same consequences.

法律责任 · PDPALEGAL DUTIES · PDPA

个人数据违规:财务处罚上限Personal-data contraventions: penalty ceilings

10%

新加坡年营业额超过1,000万新元的机构:以新加坡年营业额计算。For organisations with annual Singapore turnover above S$10 million: calculated on Singapore turnover.

S$1,000,000

其他情况:相关财务处罚上限为100万新元。Otherwise: the relevant financial penalty ceiling is S$1 million.

  • 适用前提:符合相关PDPA执法条件的数据保护违规。When relevant: data-protection contraventions meeting the applicable PDPA enforcement conditions.
  • 还可能有:整改指示;具体措施取决于事实及执法决定。Other outcomes: remedial directions; actual measures depend on the facts and decision.
特别注意:这是上限,不是固定罚款;10%不是按全球营业额计算。Important: these are ceilings, not automatic fines. The 10% is not based on worldwide turnover.
了解适用条件与处罚说明Read about conditions and penalties →
最新处罚案例 · 2025RECENT PENALTY · 2025

Marina Bay Sands

公司CompanyMarina Bay Sands Pte. Ltd.
违规原因Breach客户数据平台配置错误、迁移后检查不足,导致 665,495 人的个人资料遭未授权访问及外泄。A platform misconfiguration and inadequate post-migration checks exposed personal data belonging to 665,495 people.
处罚金额PenaltyS$315,000
查看 PDPC 最终决定Read the PDPC final decision →
不要混淆:MAS监管指引 ≠ PDPA法律处罚 ≠ 自愿AI治理框架。本页不为MAS指引编造单独罚款金额。Keep these distinct: MAS supervisory guidelines ≠ PDPA penalties ≠ voluntary AI governance frameworks. No standalone MAS-guideline fine is asserted here.

先分清:法律、指引与自愿框架Law, guidance and voluntary frameworks

法律要求LAW

PDPA:个人数据保护法PDPA: personal data protection

适用时,AI处理个人数据仍需遵守同意或适用例外、目的告知、合理保护、保留期限及跨境传输等要求。使用AI并不豁免既有义务。Where applicable, AI processing remains subject to consent or exceptions, purpose notification, protection, retention and overseas-transfer duties. AI does not remove existing obligations.

查看详细解读Read the explanation →
适用解释GUIDANCE

PDPC:AI个人数据指引PDPC: personal data in AI

针对AI推荐与决策系统解释PDPA的适用,也涉及第三方开发商可能承担的数据中介责任;它不是一套独立的“AI罚款表”。Explains PDPA application to AI recommendation and decision systems, including developers acting as data intermediaries. It is not a separate AI penalty schedule.

查看详细解读Read the explanation →
治理参考FRAMEWORK

IMDA:AI治理框架IMDA: AI governance frameworks

2020年模型AI治理框架属自愿指引;2024年生成式AI框架及2026年智能体AI框架提供治理参考。采用框架不等于取得合规认证,也不能代替法律义务。The 2020 model framework is voluntary. The 2024 generative-AI and 2026 agentic-AI frameworks offer governance references. Adoption is not compliance certification and does not replace legal duties.

查看详细解读Read the explanation →
监管期望 · 金融机构SUPERVISORY EXPECTATIONS · FINANCIAL INSTITUTIONS

第三方AI,责任不能一并外包Third-party AI does not outsource accountability

适用对象:金融机构及各种AI技术。性质:MAS监管指引,不是所有企业通用的罚款法规。Scope: financial institutions and all AI technologies. Status: MAS supervisory guidelines—not a universal fine schedule.
发布Published2026-10-07
第一阶段开始First phase starts2027-10-07
第二阶段完成期限Second phase due2028-10-07

机构需要关注什么?What should institutions focus on?

  • 有人负责:管理层监督,明确AI风险责任。Ownership: management oversight and clear accountability.
  • 知道在用什么:建立AI使用清单并评估风险。Visibility: maintain AI inventories and assess risk.
  • 持续管理:关注AI全生命周期控制和第三方保障。Ongoing controls: manage life-cycle risks and third-party assurance.
  • 风险过高要调整:无法控制在承受范围内时,考虑限制、暂停或替换第三方服务。Act on excessive risk: consider limiting, suspending or replacing services where risks remain outside appetite.
查看金融机构AI责任解读Read about AI accountability in finance →

上线AI前,先问这六件事Six questions before deploying AI

以下是BeeAI整理的实务自查建议,并非完整法定义务清单。The following is a BeeAI practical checklist, not an exhaustive list of statutory duties.

  • 谁负责?明确业务负责人、数据负责人和升级处理人。Who owns the risk? Name business, data and escalation owners.
  • 用了什么数据?核对来源、授权、用途、保存与跨境流向。What data is used? Check provenance, authority, purpose, retention and transfers.
  • 供应商承诺什么?核对数据使用、训练用途、分包、删除与事故协作。What does the vendor commit to? Check training use, subprocessors, deletion and incident cooperation.
  • 高风险决定谁复核?为重要输出设置人工检查和纠错渠道。Who reviews important decisions? Provide human checks and correction channels.
  • 出问题如何停?准备权限限制、日志、暂停机制和事件响应流程。How can it be stopped? Prepare access limits, logs, suspension and incident procedures.
  • 变化后谁复查?模型、用途或数据变化时重新评估风险。Who reassesses changes? Review risk when models, purposes or data change.

读完,做个小测试Check your understanding

6道单选题 · 完成后显示得分、正确答案与解析。不收集姓名或联系方式,答案不上传、不保存。测试不是合规认证。6 single-choice questions · Get a score, answers and explanations. No names or contact details collected; answers are not uploaded or stored. This is not compliance certification.

1. MAS这份指引主要适用于谁?Who are the MAS guidelines aimed at?
2. 金融机构发现第三方AI风险无法降到可接受范围,应该怎么办?What should an FI do if third-party AI risks cannot be reduced to an acceptable level?
3. 使用第三方AI后,金融机构可以不再负责吗?Does using third-party AI remove FI accountability?
4. 机构新加坡年营业额为2,000万新元,相关PDPA处罚上限如何理解?With S$20 million annual Singapore turnover, what is the relevant PDPA ceiling?
5. Marina Bay Sands 案例为何被处罚?Why was Marina Bay Sands penalised?
6. 采用自愿AI治理框架是否等于取得法律合规认证?Does adopting a voluntary AI framework certify legal compliance?

资料将随法规发展而变化。请以页面标注的核对日期及官方最新版本为准。Regulatory material changes. Refer to the review date and the latest official versions.