Healthcare AI Patient-Data Security Requirements

<h1>Healthcare AI Patient-Data Security Requirements</h1> <p>Singapore's Ministry of Health (<strong>MOH</strong>) states that mandatory data-security requirements apply to AI tools that process patient data, whether the tools are hosted on <strong>third-party cloud services</strong> or <strong>on-premise</strong> (<a href="https://www.moh.gov.sg/newsroom/data-security-requirements-apply-to-ai-tools-that-process-patient-data/">MOH, 4 August 2026</a>). MOH identifies the Healthcare Services Act and the Personal Data Protection Act as the legal bases for these requirements. The rule is about handling patient data in healthcare; it is not a blanket approval of every clinical AI tool or a claim that cloud hosting removes healthcare obligations.</p> <h2>Public-healthcare safeguards for AI providers</h2> <p>MOH says public healthcare institutions have adopted additional practices beyond the baseline legal requirements. AI model providers working with those institutions must give <strong>legally binding commitments that input and output data are not stored or retained</strong>, and the AI tools must be accessed from <strong>secure environments</strong> (<a href="https://www.moh.gov.sg/newsroom/data-security-requirements-apply-to-ai-tools-that-process-patient-data/">MOH, 4 August 2026</a>). These are safeguards described for public healthcare institutions and their providers; the parliamentary answer does not say that all AI vendors receive the same contract, nor that “no retention” replaces all other security, privacy or clinical-governance duties.</p> <p>The additional public-sector practice should be read as a procurement and deployment safeguard layered on top of the legal baseline. It addresses what a provider commits to do with input and output data and where the tool may be accessed; it does not establish that the model is clinically effective, that a particular product is approved for every use, or that public institutions share one universal technical architecture. The responsible institution still has to assess the proposed use, the data involved and the controls needed for that deployment (<a href="https://www.moh.gov.sg/newsroom/data-security-requirements-apply-to-ai-tools-that-process-patient-data/">MOH, 4 August 2026</a>).</p> <h2>Cloud, on-premise and interpretation</h2> <p>The cloud/on-premise distinction does not determine whether the requirements apply: MOH's answer expressly covers both deployment models. A healthcare organisation considering an AI service therefore still has to address applicable data-security duties when patient data is processed through an external cloud, just as it does for an internally hosted tool (<a href="https://www.moh.gov.sg/newsroom/data-security-requirements-apply-to-ai-tools-that-process-patient-data/">MOH, 4 August 2026</a>). The answer is a dated parliamentary clarification of existing legal requirements and public-sector practice, not a new standalone AI licensing regime or a substitute for checking the applicable Acts and institutional controls.</p> <p>In practical terms, “cloud” describes where the service is hosted, while “on-premise” describes a deployment within the organisation’s own environment; neither label answers whether patient data is being processed or what safeguards are required. A vendor’s promise not to retain data also does not by itself answer access control, secure-environment, accountability, clinical-safety or medical-device questions. Those questions must be assessed against the Healthcare Services Act, the Personal Data Protection Act, AIHGle 2.0, HSA guidance where relevant and the healthcare organisation’s own governance arrangements (<a href="https://www.moh.gov.sg/newsroom/data-security-requirements-apply-to-ai-tools-that-process-patient-data/">MOH, 4 August 2026</a>; <a href="https://www.moh.gov.sg/others/health-regulation/emerging-regulatory-policy-issues/">MOH, AIHGle 2.0 overview</a>).</p> <h2>AIHGle 2.0: lifecycle responsibilities and deployment governance</h2> <p>MOH and the Health Sciences Authority updated the <strong>Artificial Intelligence in Healthcare Guidelines (AIHGle 2.0)</strong> in March 2026. The framework builds on the 2021 version and provides recommendations and good practices alongside prevailing legislation; it is not itself a new security statute. It focuses on machine-learning and deep-learning applications used in clinical and clinical-operations work, where patient care outcomes may be affected directly or indirectly. AIHGle 2.0 does not target operational-only AI tools such as scheduling or inventory management; MOH points healthcare organisations to sector-agnostic national AI guidance for those uses. This scope boundary does not remove the separate MOH-stated data-security requirements when any AI tool processes patient data (<a href="https://isomer-user-content.by.gov.sg/3/6636a140-7986-43f6-9861-096333138d59/AIHGle_2.0.pdf">MOH and HSA, AIHGle 2.0, March 2026, sections 2.1–2.2</a>; <a href="https://www.moh.gov.sg/others/health-regulation/emerging-regulatory-policy-issues/">MOH, Emerging regulatory policy issues, updated 13 April 2026</a>).</p> <p>AIHGle 2.0 separates responsibilities among developers, healthcare organisations that deploy AI, and healthcare professionals who use it. For clinical and clinical-operations use, human oversight is mandatory: AI should assist rather than replace professional judgement, and professionals remain responsible for safe and effective care. Deployers are expected to maintain governance, risk management, quality assurance, staff readiness and incident reporting. These are guidelines for responsible practice, not a claim that every healthcare AI product has been approved or that every operational tool has identical clinical controls (<a href="https://isomer-user-content.by.gov.sg/3/6636a140-7986-43f6-9861-096333138d59/AIHGle_2.0.pdf">MOH and HSA, AIHGle 2.0, sections 4–7</a>).</p> <p>AIHGle 2.0 does not displace the Healthcare Services Act, Personal Data Protection Act or HSA’s medical-device rules where those apply. MOH describes it as a living document that will be updated as technologies develop; healthcare institutions must therefore read it together with applicable legal obligations, clinical governance, cybersecurity controls and the risk profile of the particular system. The framework’s purpose is to support safe and trusted adoption while keeping patient safety and clinical effectiveness central (<a href="https://www.moh.gov.sg/others/health-regulation/emerging-regulatory-policy-issues/">MOH, AIHGle 2.0 overview</a>).</p>

简介

医疗人工智能患者数据安全要求

新加坡卫生部(MOH)表示,无论处理患者数据的人工智能工具托管在第三方云服务还是本地环境,都必须符合强制性数据安全要求(MOH,2026 年八月 4 日)。卫生部指出,这些要求的法律依据是《医疗服务法》和《个人资料保护法》。这些规定针对医疗领域患者数据的处理;它们并非对所有临床人工智能工具的一揽子批准,也不表示使用云托管就免除了医疗领域的义务。

公立医疗机构对人工智能供应商的保障措施

卫生部表示,公立医疗机构已采取超出法律基线要求的额外做法。与这些机构合作的人工智能模型供应商必须作出具有法律约束力的承诺,保证不储存或保留输入和输出数据,而且必须在安全环境中访问人工智能工具(MOH,2026 年八月 4 日)。这些是公立医疗机构及其供应商所采用的保障措施;国会答复并未表示所有人工智能供应商都会签订相同合同,也未表示“不保留数据”可取代其他安全、隐私或临床治理责任。

应将这项额外的公共部门做法理解为叠加在法律基线之上的采购和部署保障措施。它涉及供应商承诺如何处理输入和输出数据,以及可以在何处访问工具;它并未确立模型在临床上有效、某一产品获准用于所有用途,或公立机构采用统一的技术架构。负责的机构仍须评估拟议用途、涉及的数据,以及该部署所需的控制措施(MOH,2026 年八月 4 日)。

云端、本地部署与解读

云端与本地部署的区别并不决定这些要求是否适用:卫生部的答复明确涵盖两种部署模式。因此,考虑采用人工智能服务的医疗机构,在通过外部云服务处理患者数据时,仍须履行适用的数据安全义务;内部托管的工具也一样(MOH,2026 年八月 4 日)。该答复是对现有法律要求和公共部门做法作出的有日期的国会说明,并非一项新的独立人工智能执照制度,也不能替代对适用法令和机构控制措施的核查。

在实际应用中,“云端”描述服务的托管位置,“本地部署”描述在机构自身环境中的部署方式;这两个标签都无法说明是否正在处理患者数据,也无法说明需要哪些保障措施。供应商承诺不保留数据,本身也无法回答访问控制、安全环境、问责、临床安全或医疗器械等问题。应根据《医疗服务法》、《个人资料保护法》、AIHGle 2.0、适用情况下的 HSA 指引,以及医疗机构自身的治理安排来评估这些问题(MOH,2026 年八月 4 日;MOH,AIHGle 2.0 概览)。

AIHGle 2.0:生命周期责任与部署治理

卫生部和卫生科学局于 2026 年三月更新了医疗保健人工智能指南(AIHGle 2.0)。该框架以 2021 年版本为基础,在现行法律之外提供建议和良好做法;它本身并非新的安全法令。该框架针对用于临床和临床运营工作的机器学习和深度学习应用,因为这些应用可能直接或间接影响患者护理结果。AIHGle 2.0 不针对仅用于运营的人工智能工具,例如排程或库存管理;卫生部建议医疗机构在这些用途上参考不局限于特定领域的国家人工智能指南。这一适用范围界限不会取消卫生部所述的另一项数据安全要求,即任何人工智能工具处理患者数据时都必须遵守相关要求(MOH 和 HSA,AIHGle 2.0,2026 年三月,第 2.1 至 2.2 节;MOH,新兴监管政策议题,更新于 2026 年四月 13 日)。

AIHGle 2.0 区分开发者、部署人工智能的医疗机构,以及使用人工智能的医疗专业人员各自的责任。在临床和临床运营用途上,人工监督是强制要求:人工智能应提供辅助,而非取代专业判断;专业人员仍须对安全有效的护理负责。部署方应维持治理、风险管理、质量保证、员工准备和事故通报。这些是负责任实践指南,并不表示每一种医疗人工智能产品都已获批准,也不表示所有运营工具都适用相同的临床控制措施(MOH 和 HSA,AIHGle 2.0,第 4 至 7 节)。

在适用情况下,AIHGle 2.0 不会取代《医疗服务法》、《个人资料保护法》或 HSA 的医疗器械规则。卫生部称该指南是一份会随着技术发展而更新的动态文件;因此,医疗机构必须结合适用法律义务、临床治理、网络安全控制措施,以及特定系统的风险状况来解读。该框架旨在支持安全、可信地采用人工智能,同时将患者安全和临床成效置于核心位置(MOH,AIHGle 2.0 概览)。