Azure 机器学习的 Azure 安全基线Azure security baseline for Azure Machine Learning

适用于 Microsoft Azure 机器学习的 Azure 安全基线包含可帮助你改进部署安全状况的建议。The Azure Security Baseline for Microsoft Azure Machine Learning contains recommendations that will help you improve the security posture of your deployment. 此服务的基线摘自 Azure 安全基准版本 1.0,其中提供了有关如何根据我们的最佳做法指导保护 Azure 上的云解决方案的建议。The baseline for this service is drawn from the Azure Security Benchmark version 1.0, which provides recommendations on how you can secure your cloud solutions on Azure with our best practices guidance. 有关详细信息,请参阅 Azure 安全基线概述For more information, see Azure Security Baselines overview.

网络安全性Network security

有关详细信息,请参阅 Azure 安全基线: 网络安全For more information, see the Azure Security Benchmark: Network security.

1.1:保护虚拟网络中的 Azure 资源1.1: Protect Azure resources within virtual networks

指导:Azure 机器学习依赖于其他 Azure 服务提供计算资源。Guidance: Azure Machine Learning relies on other Azure services for compute resources. 计算资源(计算目标)用于训练和部署模型。Compute resources (compute targets) are used to train and deploy models. 可以在虚拟网络中创建这些计算目标。You can create these compute targets in a virtual network. 例如,你可以使用 Azure 虚拟机器学习计算实例来训练模型,然后将模型部署到 Azure Kubernetes 服务 (AKS)。For example, you can use Azure Virtual Machine Learning compute instance to train a model and then deploy the model to Azure Kubernetes Service (AKS). 你可以通过在 Azure 虚拟网络中隔离 Azure 机器学习训练和推理作业来保护机器学习生命周期。You can secure your machine learning lifecycles by isolating Azure Machine Learning training and inference jobs within an Azure virtual network.

Azure 防火墙可用于控制对 Azure 机器学习工作区和公共 Internet 的访问。Azure Firewall can be used to control access to your Azure Machine Learning workspace and the public internet.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

1.2:监视和记录虚拟网络、子网和 NIC 的配置与流量1.2: Monitor and log the configuration and traffic of virtual networks, subnets, and NICs

指导:Azure 机器学习依赖于其他 Azure 服务提供计算资源。Guidance: Azure Machine Learning relies on other Azure services for compute resources. 将网络安全组分配给作为机器学习部署创建的网络。Assign network security groups to the networks that are created as your Machine Learning deployment.

启用网络安全组流日志,并将日志发送到 Azure 存储帐户进行审核。Enable network security group flow logs and send the logs to an Azure Storage account for auditing. 你还可以将流日志发送到 Log Analytics 工作区,然后使用流量分析来提供有关 Azure 云中流量模式的见解。You can also send the flow logs to a Log Analytics workspace and then use Traffic Analytics to provide insights into traffic patterns in your Azure cloud. 流量分析的优势包括能够可视化网络活动、识别热点和安全威胁、了解通信流模式,以及查明网络不当配置。Some advantages of Traffic Analytics are the ability to visualize network activity, identify hot spots and security threats, understand traffic flow patterns, and pinpoint network misconfigurations.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

1.3:保护关键 Web 应用程序1.3: Protect critical web applications

指导:可以启用 HTTPS 来保护与 Azure 机器学习所部署的 Web 服务的通信。Guidance: You can enable HTTPS to secure communication with web services deployed by Azure Machine Learning. Web 服务部署在 Azure Kubernetes 服务 (AKS) 或 Azure 容器实例 (ACI) 上,可保护客户端提交的数据。Web services are deployed on Azure Kubernetes Services (AKS) or Azure Container Instances (ACI) and secure the data submitted by clients. 还可以将专用 IP 与 AKS 配合使用来限制评分,使得只有虚拟网络后面的客户端才能访问 Web 服务。You can also use private IP with AKS to restrict scoring, so that only clients behind a virtual network can access the web service.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

1.4:拒绝与已知恶意的 IP 地址进行通信1.4: Deny communications with known malicious IP addresses

指导:在与机器学习实例关联的虚拟网络上启用 DDoS 防护标准,以防范分布式拒绝服务 (DDoS) 攻击。Guidance: Enable DDoS Protection Standard on the virtual networks associated with your Machine Learning instance to guard against distributed denial-of-service (DDoS) attacks. 使用 Azure 安全中心集成的威胁检测来检测与已知恶意的或未使用过的 Internet IP 地址的通信。Use Azure Security Center Integrated threat detection to detect communications with known malicious or unused Internet IP addresses.

在组织的每个网络边界上部署 Azure 防火墙,启用基于威胁情报的筛选并将其配置为针对恶意网络流量执行“发出警报并拒绝”操作。Deploy Azure Firewall at each of the organization's network boundaries with threat intelligence-based filtering enabled and configured to "Alert and deny" for malicious network traffic.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

1.5:记录网络数据包1.5: Record network packets

指导:对于在 Azure 机器学习服务中安装了合适扩展的任何 VM,可以启用网络观察程序数据包捕获来调查异常活动。Guidance: For any VMs with the proper extension installed in your Azure Machine Learning services, you can enable Network Watcher packet capture to investigate anomalous activities.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

1.6:部署基于网络的入侵检测/入侵防护系统 (IDS/IPS)1.6: Deploy network-based intrusion detection/intrusion prevention systems (IDS/IPS)

指导:在组织的每个网络边界上部署所选的防火墙解决方案,以检测并/或阻止恶意流量。Guidance: Deploy the firewall solution of your choice at each of your organization's network boundaries to detect and/or block malicious traffic.

从 Azure 市场中选择一种产品/服务,该产品/服务应支持包含有效负载检查功能的 IDS/IPS 功能。Select an offer from Azure Marketplace that supports IDS/IPS functionality with payload inspection capabilities. 如果不需要进行有效负载检查,则可以使用 Azure 防火墙威胁情报。When payload inspection is not a requirement, Azure Firewall threat intelligence can be used. 使用基于 Azure 防火墙威胁情报的筛选功能,针对进出已知恶意 IP 地址和域的流量发出警报并/或阻止该流量。Azure Firewall threat intelligence-based filtering is used to alert on and/or block traffic to and from known malicious IP addresses and domains. IP 地址和域源自 Microsoft 威胁智能源。The IP addresses and domains are sourced from the Microsoft Threat Intelligence feed.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

1.7:管理发往 Web 应用程序的流量1.7: Manage traffic to web applications

指导:不适用;此建议适用于 Azure 应用服务或计算资源上运行的 Web 应用程序。Guidance: Not applicable; this recommendation is intended for web applications running on Azure App Service or compute resources.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

1.8:最大程度地降低网络安全规则的复杂性和管理开销1.8: Minimize complexity and administrative overhead of network security rules

指导:对于需要访问你的 Azure 机器学习帐户的资源,请使用虚拟网络服务标记来定义网络安全组或 Azure 防火墙上的网络访问控制。Guidance: For resources that need access to your Azure Machine Learning account, use Virtual Network service tags to define network access controls on network security groups or Azure Firewall. 创建安全规则时,可以使用服务标记代替特定的 IP 地址。You can use service tags in place of specific IP addresses when creating security rules. 通过在规则的相应源或目标字段中指定服务标记名(例如,AzureMachineLearning),可以允许或拒绝相应服务的流量。By specifying the service tag name (for example, AzureMachineLearning) in the appropriate source or destination field of a rule, you can allow or deny the traffic for the corresponding service. Microsoft 会管理服务标记包含的地址前缀,并会在地址发生更改时自动更新服务标记。Microsoft manages the address prefixes encompassed by the service tag and automatically updates the service tag as addresses change.

Azure 机器学习服务在虚拟网络中记录其计算目标的服务标记列表,有助于最大程度地降低复杂性。你可以在网络管理中将其用作指导原则。Azure Machine Learning service documents a list of service tags for its compute targets within a virtual network that helps to minimize complexity, you can use it as guidelines in your network management.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

1.9:维护网络设备的标准安全配置1.9: Maintain standard security configurations for network devices

指导:使用 Azure Policy 为与 Azure 机器学习命名空间关联的网络资源定义和实施标准安全配置。Guidance: Define and implement standard security configurations for network resources associated with your Azure Machine Learning namespaces with Azure Policy. 在“Microsoft.MachineLearning”和“Microsoft.Network”命名空间中使用 Azure Policy 别名创建自定义策略,以审核或强制实施机器学习命名空间的网络配置。Use Azure Policy aliases in the "Microsoft.MachineLearning" and "Microsoft.Network" namespaces to create custom policies to audit or enforce the network configuration of your Machine Learning namespaces.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

1.10:阐述流量配置规则1.10: Document traffic configuration rules

指导:对与 Azure 机器学习部署关联的网络资源使用标记,以便有条理地根据分类来组织这些资源。Guidance: Use tags for network resources associated with your Azure Machine Learning deployment in order to logically organize them according to a taxonomy.

对于 Azure 机器学习虚拟网络中支持“说明”字段的资源,请使用它来记录允许进出某个网络的流量的规则。For a resource in your Azure Machine Learning virtual network that support the Description field, use it to document the rules that allow traffic to/from a network.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

1.11:使用自动化工具来监视网络资源配置和检测更改1.11: Use automated tools to monitor network resource configurations and detect changes

指导:使用 Azure 活动日志监视网络资源配置,并检测与 Azure 机器学习相关的网络资源的更改。Guidance: Use Azure Activity Log to monitor network resource configurations and detect changes for network resources related to Azure Machine Learning. 在 Azure Monitor 中创建当关键网络资源发生更改时触发的警报。Create alerts within Azure Monitor that will trigger when changes to critical network resources take place.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

日志记录和监视Logging and monitoring

有关详细信息,请参阅 Azure 安全基线: 日志记录和监视For more information, see the Azure Security Benchmark: Logging and monitoring.

2.1:使用批准的时间同步源2.1: Use approved time synchronization sources

指导:Microsoft 为日志中的时间戳维护用于 Azure 资源(例如 Azure 机器学习)的时间源。Guidance: Microsoft maintains the time source used for Azure resources such as Azure Machine Learning for timestamps in the logs.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:MicrosoftResponsibility: Microsoft

2.2:配置中心安全日志管理2.2: Configure central security log management

指导:通过 Azure Monitor 引入日志来聚合由 Azure 机器学习生成的安全数据。Guidance: Ingest logs via Azure Monitor to aggregate security data generated by Azure Machine Learning. 在 Azure Monitor 中,使用 Log Analytics 工作区来查询和执行分析,并使用 Azure 存储帐户进行长期存档存储。In Azure Monitor, use Log Analytics workspaces to query and perform analytics, and use Azure Storage accounts for long term and archival storage. 或者,可以启用数据并将其加入 Azure Sentinel 或第三方安全信息和事件管理 (SIEM)。Alternatively, you may enable, and on-board data to Azure Sentinel or a third-party Security Incident and Event Management (SIEM).

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

2.3:为 Azure 资源启用审核日志记录2.3: Enable audit logging for Azure resources

指导:在 Azure 资源上启用诊断设置,以访问审核日志、安全日志和诊断日志。Guidance: Enable diagnostic settings on Azure resources for access to audit, security, and diagnostic logs. 活动日志自动可用,包括事件源、日期、用户、时间戳、源地址、目标地址和其他有用元素。Activity logs, which are automatically available, include event source, date, user, timestamp, source addresses, destination addresses, and other useful elements.

你还可以将机器学习服务操作日志进行关联,以提高安全性与合规性。You can also correlate Machine Learning service operation logs for security and compliance purposes.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

2.4:从操作系统收集安全日志2.4: Collect security logs from operating systems

指导:如果计算资源归 Microsoft 所有,则 Microsoft 负责收集并监视它。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for collecting and monitoring it.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的任何计算资源,请使用 Azure 安全中心来监视操作系统。For any compute resources that are owned by your organization, use Azure Security Center to monitor the operating system.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:共享Responsibility: Shared

2.5:配置安全日志存储保留期2.5: Configure security log storage retention

指导:在 Azure Monitor 中,根据组织的合规性制度,为与你的 Azure 机器学习实例关联的 Log Analytics 工作区设置日志保留期。Guidance: In Azure Monitor, set the log retention period for Log Analytics workspaces associated with your Azure Machine Learning instances according to your organization's compliance regulations.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

2.6:监视和审查日志2.6: Monitor and review Logs

指导:分析和监视日志中的异常行为,并定期审查来自 Azure 机器学习的结果。Guidance: Analyze and monitor logs for anomalous behavior and regularly review the results from your Azure Machine Learning. 使用 Azure Monitor 和 Log Analytics 工作区查看日志并对日志数据执行查询。Use Azure Monitor and a Log Analytics workspace to review logs and perform queries on log data.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

2.7:针对异常活动启用警报2.7: Enable alerts for anomalous activities

指导:在 Azure Monitor 中,配置与活动日志和机器学习诊断设置中的 Azure 机器学习相关的日志,以将日志发送到 Log Analytics 工作区供查询或发送到某个存储帐户进行长期存档存储。Guidance: In Azure Monitor, configure logs related to Azure Machine Learning within the Activity Log, and Machine Learning diagnostic settings to send logs into a Log Analytics workspace to be queried or into a storage account for long-term archival storage. 使用 Log Analytics 工作区针对安全日志和事件中的异常活动创建警报。Use Log Analytics workspace to create alerts for anomalous activity found in security logs and events.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

2.8:集中管理反恶意软件日志记录2.8: Centralize anti-malware logging

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的 Antimalware 部署。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for Antimalware deployment of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请为 Azure 云服务和虚拟机的 Microsoft Antimalware 启用反恶意软件事件收集。For compute resources that are owned by your organization, enable antimalware event collection for Microsoft Antimalware for Azure Cloud Services and Virtual Machines.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

2.9:启用 DNS 查询日志记录2.9: Enable DNS query logging

指导:不适用;Azure 机器学习不会处理或生成与 DNS 相关的日志。Guidance: Not applicable; Azure Machine Learning does not process or produce DNS-related logs.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

2.10:启用命令行审核日志记录2.10: Enable command-line audit logging

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 安全中心为 Azure 虚拟机启用安全事件日志监视。For compute resources are owned by your organization, use Azure Security Center to enable security event log monitoring for Azure virtual machines. 如果启用了自动预配,则 Azure 安全中心会在所有受支持的 Azure VM 以及任何新建的 Azure VM 中预配 Log Analytics 代理。Azure Security Center provisions the Log Analytics agent on all supported Azure VMs, and any new ones that are created if automatic provisioning is enabled. 你也可以手动安装代理。Or you can install the agent manually. 该代理可启用进程创建事件 4688 和事件 4688 内的 CommandLine 字段。The agent enables the process creation event 4688 and the commandline field inside event 4688. VM 上创建的新进程由事件日志记录,由安全中心的检测服务监视。New processes created on the VM are recorded by event log and monitored by Security Center's detection services.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

标识和访问控制Identity and access control

有关详细信息,请参阅 Azure 安全基线: 标识和访问控制For more information, see the Azure Security Benchmark: Identity and access control.

3.1:维护管理帐户的清单3.1: Maintain an inventory of administrative accounts

指导:可以使用 Azure 门户中资源的“标识和访问管理”选项卡配置基于角色的访问控制 (RBAC),并维护有关 Azure 机器学习资源的库存。Guidance: You can use the Identity and Access Management tab for a resource in the Azure portal to configure role-based access control (RBAC) and maintain inventory on Azure Machine Learning resources. 角色将应用到 Active Directory 中的用户、组、服务主体和托管标识。The roles are applied to users, groups, service principals, and managed identities in Active Directory. 对于个人和组,可使用内置角色或自定义角色。You can use built-in roles or custom roles for individuals and groups.

Azure 机器学习为 Azure 机器学习中的常见管理方案提供了内置的 RBAC。Azure Machine Learning provides built-in RBAC for common management scenarios in Azure Machine Learning. 在 Azure Active Directory (Azure AD) 中创建了配置文件的个人可将这些 RBAC 角色分配给用户、组、服务主体或托管标识,以授予或拒绝对资源和 Azure 机器学习资源操作的访问权限。An individual who has a profile in Azure Active Directory (Azure AD) can assign these RBAC roles to users, groups, service principals, or managed identities to grant or deny access to resources and operations on Azure Machine Learning resources.

还可以使用 Azure AD PowerShell 模块执行即席查询,以发现属于管理组的成员的帐户。You can also use the Azure AD PowerShell module to perform adhoc queries to discover accounts that are members of administrative groups.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

3.2:在适用的情况下更改默认密码3.2: Change default passwords where applicable

指导:对机器学习资源的访问管理是通过 Azure Active Directory (Azure AD) 控制的。Guidance: Access management to Machine Learning resources is controlled through Azure Active Directory (Azure AD). Azure AD 没有默认密码。Azure AD does not have the concept of default passwords.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.3:使用专用管理帐户3.3: Use dedicated administrative accounts

指导:当创建新的工作区时,Azure 机器学习附带了三个默认角色,并创建与所有者帐户的使用相关的标准操作规程。Guidance: Azure Machine Learning comes with three default roles when a new workspace is created, creating standard operating procedures around the use of owner accounts.

还可以通过使用 Azure AD Privileged Identity Management 和 Azure 资源管理器来启用对管理帐户的即时访问权限。You can also enable a just-in-time access to administrative accounts by using Azure AD Privileged Identity Management and Azure Resource Manager.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

3.4:将单一登录 (SSO) 与 Azure Active Directory 配合使用3.4: Use single sign-on (SSO) with Azure Active Directory

指导:机器学习与 Azure Active Directory 集成。请使用 Azure Active Directory SSO,而不是为每个服务配置单个独立凭据。Guidance: Machine Learning is integrated with Azure Active Directory, use Azure Active Directory SSO instead of configuring individual stand-alone credentials per-service. 请使用 Azure 安全中心标识和访问建议。Use Azure Security Center identity and access recommendations.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.5:对所有基于 Azure Active Directory 的访问使用多重身份验证3.5: Use multi-factor authentication for all Azure Active Directory based access

指导:启用 Azure Active Directory 多重身份验证,并遵循 Azure 安全中心标识和访问建议。Guidance: Enable Azure Active Directory Multi-Factor Authentication and follow Azure Security Center identity and access recommendations.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

3.6:对所有管理任务使用专用计算机(特权访问工作站)3.6: Use dedicated machines (Privileged Access Workstations) for all administrative tasks

指导:对于需要提升的权限的管理任务,请使用安全的 Azure 托管工作站(也称为特权访问工作站,简称 PAW)。Guidance: Use a secure, Azure-managed workstation (also known as a Privileged Access Workstation, or PAW) for administrative tasks that require elevated privileges.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.7:记录来自管理帐户的可疑活动并对其发出警报3.7: Log and alert on suspicious activities from administrative accounts

指导:使用 Azure Active Directory 安全报告和监视来检测环境中何时发生可疑活动或不安全的活动。Guidance: Use Azure Active Directory security reports and monitoring to detect when suspicious or unsafe activity occurs in the environment. 使用 Azure 安全中心监视标识和访问活动。Use Azure Security Center to monitor identity and access activity.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

3.8:仅从批准的位置管理 Azure 资源3.8: Manage Azure resources only from approved locations

指导:使用 Azure AD 命名位置,仅允许从 IP 地址范围或国家/地区的特定逻辑分组进行访问。Guidance: Use Azure AD named locations to allow access only from specific logical groupings of IP address ranges or countries/regions.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.9:使用 Azure Active Directory3.9: Use Azure Active Directory

指导:使用 Azure Active Directory (Azure AD) 作为中心身份验证和授权系统。Guidance: Use Azure Active Directory (Azure AD) as the central authentication and authorization system. Azure AD 通过对静态数据和传输中数据使用强加密来保护数据。Azure AD protects data by using strong encryption for data at rest and in transit. Azure AD 还会对用户凭据进行加盐、哈希处理和安全存储操作。Azure AD also salts, hashes, and securely stores user credentials.

在 Azure 中,角色访问的作用域可以限定为多个级别。Role access can be scoped to multiple levels in Azure. 对于机器学习,可以在工作区级别管理角色,例如,你可能拥有某个工作区的所有者访问权限,但没有该工作区所在的资源组的所有者访问权限。For Machine Learning, roles can be managed at workspace level, for example, you have owner access to a workspace may not have owner access to the resource group that contains the workspace. 这将提供更精细的访问控制,以在同一资源组中分隔角色。This provides more granular access controls to separate roles within the same resource group.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.10:定期审查和协调用户访问3.10: Regularly review and reconcile user access

指南:Azure AD 提供日志来帮助发现过时的帐户。Guidance: Azure AD provides logs to help discover stale accounts. 此外,请使用 Azure AD 标识和访问评审来有效管理组成员身份、对企业应用程序的访问以及角色分配。In addition, use Azure AD identity and access reviews to efficiently manage group memberships, access to enterprise applications, and role assignments. 可以定期评审用户的访问权限,确保只有适当的用户才持续拥有访问权限。User access can be reviewed on a regular basis to make sure only the right users have continued access.

当环境中出现可疑或不安全的活动时,可使用 Azure Active Directory (Azure AD) Privileged Identity Management (PIM) 生成日志和警报。Use Azure Active Directory (Azure AD) Privileged Identity Management (PIM) for generation of logs and alerts when suspicious or unsafe activity occurs in the environment.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

3.11:监视尝试访问已停用凭据的行为3.11: Monitor attempts to access deactivated credentials

指导:你有权访问 Azure AD 登录活动、审核和风险事件日志源,因此可以与任何 SIEM/监视工具集成。Guidance: You have access to Azure AD sign-in activity, audit, and risk event log sources, which allow you to integrate with any SIEM/monitoring tool.

可以通过为 Azure AD 用户帐户创建诊断设置,并将审核日志和登录日志发送到 Log Analytics 工作区,来简化此过程。You can streamline this process by creating diagnostic settings for Azure AD user accounts and sending the audit logs and sign-in logs to a Log Analytics workspace. 你可以在 Log Analytics 工作区中配置所需的警报。You can configure desired alerts within Log Analytics workspace.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.12:针对帐户登录行为偏差发出警报3.12: Alert on account login behavior deviation

指导:使用 Azure AD 标识保护功能来配置对检测到的与用户标识相关的可疑操作的自动响应。Guidance: Use Azure AD Identity Protection features to configure automated responses to detected suspicious actions related to user identities. 还可将数据引入 Azure Sentinel 以做进一步调查。You can also ingest data into Azure Sentinel for further investigation.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

3.13:在支持场合下为 Microsoft 提供对相关客户数据的访问权限3.13: Provide Microsoft with access to relevant customer data during support scenarios

指导:不适用;Azure 机器学习服务不支持客户密码箱。Guidance: Not applicable; Azure Machine Learning service doesn’t support customer lockbox.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

数据保护Data protection

有关详细信息,请参阅 Azure 安全基线: 数据保护For more information, see the Azure Security Benchmark: Data protection.

4.1:维护敏感信息的清单4.1: Maintain an inventory of sensitive Information

指导:使用标记可以帮助跟踪存储或处理敏感信息的 Azure 资源。Guidance: Use tags to assist in tracking Azure resources that store or process sensitive information.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.2:隔离存储或处理敏感信息的系统4.2: Isolate systems storing or processing sensitive information

指导:使用单独的订阅和管理组对各个安全域(如环境类型和数据敏感度级别)实现隔离。Guidance: Implement isolation using separate subscriptions and management groups for individual security domains such as environment type and data sensitivity level. 你可以限制对应用程序和企业环境所需 Azure 资源的访问级别。You can restrict the level of access to your Azure resources that your applications and enterprise environments demand. 可以通过 Azure RBAC 来控制对 Azure 资源的访问。You can control access to Azure resources via Azure RBAC.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.3:监视和阻止未经授权的敏感信息传输4.3: Monitor and block unauthorized transfer of sensitive information

指导:利用 Azure 市场中有关网络外围的第三方解决方案,监视并阻止敏感信息的未授权传输,同时提醒信息安全专业人员。Guidance: Use a third-party solution from Azure Marketplace in network perimeters to monitor for unauthorized transfer of sensitive information and block such transfers while alerting information security professionals.

对于 Microsoft 管理的基础平台,Microsoft 会将所有客户内容视为敏感数据,全方位防范客户数据丢失和泄露。For the underlying platform, which is managed by Microsoft, Microsoft treats all customer content as sensitive and guards against customer data loss and exposure. 为了确保 Azure 中的客户数据保持安全,Microsoft 实施并维护了一套可靠的数据保护控制措施和功能。To ensure customer data within Azure remains secure, Microsoft has implemented and maintains a suite of robust data protection controls and capabilities.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.4:加密传输中的所有敏感信息4.4: Encrypt all sensitive information in transit

指导:通过 Azure 机器学习部署的 Web 服务仅支持对数据强制实施传输中加密的 TLS 版本 1.2。Guidance: Web services deployed through Azure Machine Learning only support TLS version 1.2 that enforces data encryption in transit.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.5:使用有效的发现工具识别敏感数据4.5: Use an active discovery tool to identify sensitive data

指导:数据标识、分类和丢失防护功能尚不适用于 Azure 机器学习。Guidance: Data identification, classification, and loss prevention features are not yet available for Azure Machine Learning. 可以根据合规性需要实施第三方解决方案。Implement a third-party solution if necessary for compliance purposes.

对于由 Microsoft 管理的基础平台,Microsoft 会将所有客户内容都视为敏感信息,竭尽全力防范客户数据丢失和泄露。For the underlying platform, which is managed by Microsoft, Microsoft treats all customer content as sensitive and goes to great lengths to guard against customer data loss and exposure. 为了确保 Azure 中的客户数据保持安全,Microsoft 实施并维护了一套可靠的数据保护控制措施和功能。To ensure customer data within Azure remains secure, Microsoft has implemented and maintains a suite of robust data protection controls and capabilities.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.6:使用 Azure RBAC 管理对资源的访问4.6: Use Azure RBAC to manage access to resources

指导:Azure 机器学习支持使用 Azure Active Directory (Azure AD) 授权对机器学习资源的请求。Guidance: Azure Machine Learning supports using Azure Active Directory (Azure AD) to authorize requests to Machine Learning resources. 可以通过 Azure AD 使用 Azure 基于角色的访问控制 (RBAC) 授予对安全主体的访问权限,该安全主体可能是用户,也可能是应用程序服务主体。With Azure AD, you can use Azure role-based access control (RBAC) to grant permissions to a security principal, which may be a user, or an application service principal.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.7:使用基于主机的数据丢失防护来强制实施访问控制4.7: Use host-based data loss prevention to enforce access control

指导:不适用;此项指导适用于计算资源。Guidance: Not applicable; this guideline is intended for compute resources.

Microsoft 会管理机器学习的底层基础结构,并实施了严格的控制措施来防止客户数据丢失或泄露。Microsoft manages the underlying infrastructure for Machine Learning and has implemented strict controls to prevent the loss or exposure of customer data.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:MicrosoftResponsibility: Microsoft

4.8:静态加密敏感信息4.8: Encrypt sensitive information at rest

指导:Azure 机器学习在绑定到 Azure 机器学习工作区和订阅的 Azure Blob 存储帐户中存储快照、输出与日志。Guidance: Azure Machine Learning stores snapshots, output, and logs in the Azure Blob storage account that's tied to the Azure Machine Learning workspace and your subscription. Azure Blob 存储中存储的所有数据已通过 Microsoft 管理的密钥静态加密。All the data stored in Azure Blob storage is encrypted at rest with Microsoft-managed keys. 在机器学习服务中,还可以使用你自己的密钥加密 Azure Blob 存储中存储的数据。You can also encrypt data stored in Azure Blob storage with your own keys in Machine Learning service.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

4.9:记录对关键 Azure 资源的更改并对此类更改发出警报4.9: Log and alert on changes to critical Azure resources

指导:将 Azure Monitor 与 Azure 活动日志结合使用,以创建在 Azure 机器学习的生产实例和其他关键资源或相关资源发生更改时发出的警报。Guidance: Use Azure Monitor with the Azure Activity log to create alerts for when changes take place to production instances of Azure Machine Learning and other critical or related resources.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

漏洞管理Vulnerability management

有关详细信息,请参阅 Azure 安全基线: 漏洞管理For more information, see the Azure Security Benchmark: Vulnerability management.

5.1:运行自动漏洞扫描工具5.1: Run automated vulnerability scanning tools

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的漏洞管理。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for vulnerability management of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请遵循 Azure 安全中心提供的关于在 Azure 虚拟机、容器映像和 SQL 服务器上执行漏洞评估的建议。For compute resources that are owned by your organization, follow the recommendations from Azure Security Center for performing vulnerability assessments on your Azure virtual machines, container images, and SQL servers.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

5.2:部署自动操作系统修补管理解决方案5.2: Deploy automated operating system patch management solution

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的补丁管理。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for patch management of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的任何计算资源,请使用 Azure 自动化更新管理确保在 Windows 和 Linux VM 上安装最新的安全更新。For any compute resources that are owned by your organization, use Azure Automation Update Management to ensure that the most recent security updates are installed on your Windows and Linux VMs. 对于 Windows 虚拟机,请确保已启用 Windows 更新并将其设置为自动更新。For Windows VMs, ensure Windows Update has been enabled and set to update automatically.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

5.3:为第三方软件部署自动化补丁管理解决方案5.3: Deploy an automated patch management solution for third-party software titles

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用第三方补丁管理解决方案。For compute resources that are owned by your organization, use a third-party patch management solution. 已在其环境中使用 Configuration Manager 的客户还可以使用 System Center Updates Publisher,以便将自定义更新发布到 Windows Server 更新服务中。Customers already using Configuration Manager in their environment can also use System Center Updates Publisher, allowing them to publish custom updates into Windows Server Update Service. 这样更新管理就可以通过第三方软件来修补使用 Configuration Manager 作为其更新存储库的计算机。This allows Update Management to patch machines that use Configuration Manager as their update repository with third-party software.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

5.4:比较连续进行的漏洞扫描5.4: Compare back-to-back vulnerability scans

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请遵循 Azure 安全中心提供的关于在 Azure 虚拟机、容器映像和 SQL 服务器上执行漏洞评估的建议。For compute resources that are owned by your organization, follow recommendations from Azure Security Center for performing vulnerability assessments on your Azure virtual machines, container images, and SQL servers. 以一致的间隔导出扫描结果,并将结果与以前的扫描进行比较以验证漏洞是否已修复。Export scan results at consistent intervals and compare the results with previous scans to verify that vulnerabilities have been remediated. 使用 Azure 安全中心建议的漏洞管理建议时,可以转到选定解决方案的门户查看历史扫描数据。When using vulnerability management recommendations suggested by Azure Security Center, you can pivot into the selected solution's portal to view historical scan data.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

5.5:使用风险评级过程来确定已发现漏洞的修正措施的优先级5.5: Use a risk-rating process to prioritize the remediation of discovered vulnerabilities

指导:不适用;此项指导适用于计算资源。Guidance: Not applicable; this guideline is intended for compute resources.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

库存和资产管理Inventory and asset management

有关详细信息,请参阅 Azure 安全基线: 清单和资产管理For more information, see the Azure Security Benchmark: Inventory and asset management.

6.1:使用自动化资产发现解决方案6.1: Use automated asset discovery solution

指导:使用 Azure Resource Graph 查询和发现订阅中的资源(例如计算、存储、网络、端口和协议等)。Guidance: Use Azure Resource Graph to query for and discover resources (such as compute, storage, network, ports, and protocols etc.) in your subscriptions. 确保租户中具有适当的(读取)权限,并枚举所有 Azure 订阅以及订阅中的资源。Ensure appropriate (read) permissions in your tenant and enumerate all Azure subscriptions as well as resources in your subscriptions.

尽管可以通过 Azure Resource Graph 浏览器发现经典 Azure 资源,但我们强烈建议你今后创建并使用 Azure 资源管理器资源。Although classic Azure resources can be discovered via Azure Resource Graph Explorer, it is highly recommended to create and use Azure Resource Manager resources going forward.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.2:维护资产元数据6.2: Maintain asset metadata

指导:将标记应用于 Azure 资源,添加元数据以便根据分类有条理地进行组织。Guidance: Apply tags to Azure resources, adding metadata to logically organize according into a taxonomy.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.3:删除未经授权的 Azure 资源6.3: Delete unauthorized Azure resources

指导:在适用的情况下,请使用标记、管理组和单独的订阅来组织和跟踪资产。Guidance: Use tagging, management groups, and separate subscriptions where appropriate, to organize and track assets. 定期核对清单,确保及时地从订阅中删除未经授权的资源。Reconcile inventory on a regular basis and ensure unauthorized resources are deleted from the subscription in a timely manner.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.4:定义并维护已批准 Azure 资源的清单6.4: Define and maintain an inventory of approved Azure resources

指导:根据组织需求,创建已获批 Azure 资源以及已获批用于计算资源的软件的清单。Guidance: Create an inventory of approved Azure resources and approved software for compute resources as per your organizational needs.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.5:监视未批准的 Azure 资源6.5: Monitor for unapproved Azure resources

指导:在 Azure Policy 中使用以下内置策略定义,对可以在客户订阅中创建的资源类型施加限制:Guidance: Use Azure Policy to put restrictions on the type of resources that can be created in customer subscriptions using the following built-in policy definitions:

  • 不允许的资源类型Not allowed resource types
  • 允许的资源类型Allowed resource types

此外,请使用 Azure Resource Graph 来查询/发现订阅中的资源。In addition, use the Azure Resource Graph to query/discover resources within the subscriptions.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.6:监视计算资源中未批准的软件应用程序6.6: Monitor for unapproved software applications within compute resources

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 自动化更改跟踪和库存来自动收集 Windows 和 Linux VM 中的库存信息。For compute resources that are owned by your organization, use Azure Automation Change Tracking and Inventory to automate the collection of inventory information from your Windows and Linux VMs. 可从 Azure 门户获得软件名称、版本、发布者和刷新时间。Software name, version, publisher, and refresh time are available from the Azure portal. 若要获取软件安装日期和其他信息,请启用来宾级诊断,并将 Windows 事件日志定向到 Log Analytics 工作区。To get the software installation date and other information, enable guest-level diagnostics and direct the Windows Event Logs to Log Analytics workspace.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

6.7:删除未批准的 Azure 资源和软件应用程序6.7: Remove unapproved Azure resources and software applications

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 安全中心的文件完整性监视 (FIM) 来识别 VM 上安装的所有软件。For compute resources that are owned by your organization, use Azure Security Center's File Integrity Monitoring (FIM) to identify all software installed on VMs. 从 Linux 和 Windows VM 收集库存时,可以改用或与 FIM 一起使用的另一个选项是 Azure 自动化更改跟踪和库存。Another option that can be used instead of or in conjunction with FIM is Azure Automation Change Tracking and Inventory to collect inventory from your Linux and Windows VMs.

可以实现自己的未授权软件删除过程。You can implement your own process for removing unauthorized software. 还可以使用第三方解决方案来识别未获批软件。You can also use a third-party solution to identify unapproved software.

如果不再需要 Azure 资源,请将其删除。Remove Azure resources when they are no longer needed.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:共享Responsibility: Shared

6.8:仅使用已批准的应用程序6.8: Use only approved applications

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 安全中心自适应应用程序控制确保仅执行已授权软件,并阻止所有未授权软件在 Azure 虚拟机上执行。For compute resources that are owned by your organization, use Azure Security Center adaptive application controls to ensure that only authorized software executes and all unauthorized software is blocked from executing on Azure Virtual Machines.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:共享Responsibility: Shared

6.9:仅使用已批准的 Azure 服务6.9: Use only approved Azure services

指导:在 Azure Policy 中使用以下内置策略定义,对可以在客户订阅中创建的资源类型施加限制:Guidance: Use Azure Policy to put restrictions on the type of resources that can be created in customer subscriptions using the following built-in policy definitions:

  • 不允许的资源类型Not allowed resource types
  • 允许的资源类型Allowed resource types

此外,请使用 Azure Resource Graph 来查询并发现订阅中的资源。In addition, use the Azure Resource Graph to query for and discover resources in the subscriptions.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.10:维护已获批软件的清单6.10: Maintain an inventory of approved software titles

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的任何计算资源,请使用 Azure 安全中心自适应应用程序控制指定规则可能适用或不适用的文件类型。For any compute resources that are owned by your organization, use Azure Security Center adaptive application controls to specify which file types a rule may or may not apply to.

如果自适应应用程序控制不符合要求,请实施第三方解决方案。Implement a third-party solution if adaptive application controls don't meet the requirement.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

6.11:限制用户与 Azure 资源管理器进行交互的能力6.11: Limit users' ability to interact with Azure Resource Manager

指导:通过为“Microsoft Azure 管理”应用配置“阻止访问”,使用 Azure AD 条件访问来限制用户与 Azure 资源管理器交互的能力。Guidance: Use Azure AD Conditional Access to limit users' ability to interact with Azure Resources Manager by configuring "Block access" for the "Microsoft Azure Management" App.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

6.12:限制用户在计算资源中执行脚本的能力6.12: Limit users' ability to execute scripts in compute resources

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,根据脚本的类型,可以使用特定于操作系统的配置或第三方资源来限制用户在 Azure 计算资源中执行脚本的能力。For compute resources that are owned by your organization, depending on the type of scripts, you can use operating system-specific configurations or third-party resources to limit users' ability to execute scripts in Azure compute resources. 你还可以利用 Azure 安全中心自适应应用程序控制来确保仅执行已授权软件,并阻止所有未授权软件在 Azure 虚拟机上执行。You can also use Azure Security Center adaptive application controls to ensure that only authorized software executes and all unauthorized software is blocked from executing on Azure Virtual Machines.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

6.13:以物理或逻辑方式隔离高风险应用程序6.13: Physically or logically segregate high risk applications

指导:不适用;此建议适用于 Azure 应用服务或计算资源上运行的 Web 应用程序。Guidance: Not applicable; this recommendation is intended for web applications running on Azure App Service or compute resources.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

安全配置Secure configuration

有关详细信息,请参阅 Azure 安全基线: 安全配置For more information, see the Azure Security Benchmark: Secure configuration.

7.1:为所有 Azure 资源建立安全配置7.1: Establish secure configurations for all Azure resources

指导:使用 Azure Policy 为 Azure 机器学习服务定义和实施标准安全配置。Guidance: Define and implement standard security configurations for your Azure Machine Learning service with Azure Policy. 在“Microsoft.MachineLearning”命名空间中使用 Azure Policy 别名创建自定义策略,以审核或强制实施 Azure 机器学习服务的配置。Use Azure Policy aliases in the "Microsoft.MachineLearning" namespace to create custom policies to audit or enforce the configuration of your Azure Machine Learning services.

Azure 资源管理器能够以 JavaScript 对象表示法 (JSON) 导出模板,应该对其进行检查,以确保配置满足组织的安全要求。Azure Resource Manager has the ability to export the template in JavaScript Object Notation (JSON), which should be reviewed to ensure that the configurations meet the security requirements for your organization.

还可以使用来自 Azure 安全中心的建议作为 Azure 资源的安全配置基线。You can also use the recommendations from Azure Security Center as a secure configuration baseline for your Azure resources.

Azure 机器学习完全支持用于跟踪工作的 Git 存储库;你可以将存储库直接克隆到共享工作区文件系统上,在本地工作站上使用 Git,并确保将安全的配置作为机器学习环境的一部分应用于代码资源。Azure Machine Learning fully supports Git repositories for tracking work; you can clone repositories directly onto your shared workspace file system, use Git on your local workstation, and make sure secure configurations apply to code resources as part of your Machine Learning environment.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

7.2:建立安全的操作系统配置7.2: Establish secure operating system configurations

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的操作系统安全配置。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for operating system secure configurations of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 安全中心建议来维护所有计算资源上的安全配置。For compute resources that are owned by your organization, use Azure Security Center recommendations to maintain security configurations on all compute resources. 此外,你可以使用自定义操作系统映像或 Azure Automation State Configuration 来建立组织所需的操作系统的安全配置。Additionally, you can use custom operating system images or Azure Automation State configuration to establish the security configuration of the operating system required by your organization.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

7.3:维护安全的 Azure 资源配置7.3: Maintain secure Azure resource configurations

指南:使用 Azure Policy“[拒绝]”和“[不存在则部署]”对不同的 Azure 资源强制实施安全设置。Guidance: Use Azure Policy [deny] and [deploy if not exist] to enforce secure settings across your Azure resources. 此外,你可以使用 Azure 资源管理器模板维护组织所需的 Azure 资源的安全配置。In addition, you can use Azure Resource Manager templates to maintain the security configuration of your Azure resources required by your organization.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

7.4:维护安全的操作系统配置7.4: Maintain secure operating system configurations

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的操作系统安全配置。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for operating system secure configurations of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请遵循 Azure 安全中心提供的关于在 Azure 计算资源上执行漏洞评估的建议。For compute resources that are owned by your organization, follow recommendations from Azure Security Center on performing vulnerability assessments on your Azure compute resources. 此外,你可以使用 Azure 资源管理器模板、自定义操作系统映像或 Azure Automation State Configuration 来维护组织所需的操作系统的安全配置。In addition, you may use Azure Resource Manager templates, custom operating system images, or Azure Automation State Configuration to maintain the security configuration of the operating system required by your organization. 结合 Azure Automation State Configuration,Microsoft 虚拟机模板可能有助于满足和维护安全要求。The Microsoft virtual machine templates combined with the Azure Automation State Configuration may assist in meeting and maintaining the security requirements.

注意,由 Microsoft 发布的 Azure 市场虚拟机映像由 Microsoft 管理和维护。Note that Azure Marketplace virtual machine Images published by Microsoft are managed and maintained by Microsoft.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

7.5:安全存储 Azure 资源的配置7.5: Securely store configuration of Azure resources

指导:如果对机器学习或相关资源使用自定义 Azure Policy 定义,请使用 Azure Repos 安全地存储和管理你的代码。Guidance: If using custom Azure Policy definitions for your Machine Learning or related resources, use Azure Repos to securely store and manage your code.

Azure 机器学习完全支持用于跟踪工作的 Git 存储库;你可以将存储库直接克隆到共享工作区文件系统上,在本地工作站上使用 Git,并确保将安全的配置作为机器学习环境的一部分应用于代码资源。Azure Machine Learning fully supports Git repositories for tracking work; you can clone repositories directly onto your shared workspace file system, use Git on your local workstation, and make sure secure configurations apply to code resources as part of your Machine Learning environment.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

7.6:安全存储自定义操作系统映像7.6: Securely store custom operating system images

指导:Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Guidance: Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 基于角色的访问控制 (RBAC) 来确保只有经过授权的用户才能访问你的自定义映像。For compute resources that are owned by your organization, use Azure role-based access control (RBAC) to ensure that only authorized users can access your custom images. 使用 Azure 共享映像库,可以将映像共享给组织内的不同用户、服务主体或 Azure AD 组。Use an Azure Shared Image Gallery you can share your images to different users, service principals, or Azure AD groups within your organization. 将容器映像存储在 Azure 容器注册表中,并使用 RBAC 来确保只有经过授权的用户才能进行访问。Store container images in Azure Container Registry and use RBAC to ensure that only authorized users have access.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:不适用Responsibility: Not Applicable

7.7:部署 Azure 资源的配置管理工具7.7: Deploy configuration management tools for Azure resources

指导:在“Microsoft.MachineLearning”命名空间中使用 Azure Policy 别名创建自定义策略,以审核、强制实施系统配置并对其发出警报。Guidance: Use Azure Policy aliases in the "Microsoft.MachineLearning" namespace to create custom policies to alert, audit, and enforce system configurations. 另外,开发一个用于管理策略例外的流程和管道。Additionally, develop a process and pipeline for managing policy exceptions.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

7.8:部署操作系统的配置管理工具7.8: Deploy configuration management tools for operating systems

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责 Azure 机器学习服务的安全配置部署。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for secure configuration deployment of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请为任何云或本地数据中心内的 Desired State Configuration (DSC) 节点使用 Azure Automation State Configuration。For compute resources that are owned by your organization, use Azure Automation State Configuration for Desired State Configuration (DSC) nodes in any cloud or on-premises datacenter. 可以轻松登记计算机、为其分配声明性配置并查看显示每台计算机是否符合指定的所需状态的报告。You can easily onboard machines, assign them declarative configurations, and view reports showing each machine's compliance to the desired state you specified.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

7.9:为 Azure 资源实施自动配置监视7.9: Implement automated configuration monitoring for Azure resources

指导:使用 Azure 安全中心对 Azure 资源执行基线扫描。Guidance: Use Azure Security Center to perform baseline scans for your Azure Resources. 此外,使用 Azure Policy 警告和审核 Azure 资源配置。Additionally, use Azure Policy to alert and audit Azure resource configurations.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

7.10:为操作系统实施自动配置监视7.10: Implement automated configuration monitoring for operating systems

指导:如果计算资源由 Microsoft 拥有,则 Microsoft 负责执行 Azure 机器学习服务的自动化安全配置监视。Guidance: If the compute resource is owned by Microsoft, then Microsoft is responsible for automated secure configuration monitoring of Azure Machine Learning service.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用 Azure 安全中心的“计算和应用”,按照适用于 VM 和服务器以及容器的建议进行操作。For compute resources that are owned by your organization, use Azure Security Center Compute & Apps and follow the recommendations for VMs and servers, and containers.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

7.11:安全管理 Azure 机密7.11: Manage Azure secrets securely

指导:将托管标识与 Azure Key Vault 结合使用,以简化云应用程序的机密管理。Guidance: Use managed identities in conjunction with Azure Key Vault to simplify secret management for your cloud applications.

Azure 机器学习支持使用客户管理的密钥进行数据存储加密,你需要管理密钥轮换并废除每个组织的安全和合规性要求。Azure Machine Learning supports data store encryption with customer-managed keys, you need to manage key rotate and revoke per organization security and compliance requirements.

使用 Azure Key Vault 将机密安全地传递到远程运行,而不是在训练脚本中将其以明文形式传递。Use Azure Key Vault to pass secrets to remote runs securely instead of cleartext in your training scripts.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

7.12:安全自动管理标识7.12: Manage identities securely and automatically

指导:Azure 机器学习支持内置角色,同时还支持创建自定义角色的功能。Guidance: Azure Machine Learning supports both built-in roles and the ability to create custom roles. 使用托管标识在 Azure AD 中为 Azure 服务提供自动托管标识。Use managed identities to provide Azure services with an automatically managed identity in Azure AD. 使用托管标识可以向支持 Azure AD 身份验证的任何服务(包括 Key Vault)进行身份验证,无需在代码中放入任何凭据。Managed identities allow you to authenticate to any service that supports Azure AD authentication, including Key Vault, without any credentials in your code.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

7.13:消除意外的凭据透露7.13: Eliminate unintended credential exposure

指南:实施凭据扫描程序来识别代码中的凭据。Guidance: Implement Credential Scanner to identify credentials within code. 凭据扫描程序还会建议将发现的凭据转移到更安全的位置,例如 Azure Key Vault。Credential Scanner will also encourage moving discovered credentials to more secure locations such as Azure Key Vault.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

恶意软件防护Malware defense

有关详细信息,请参阅 Azure 安全基线: 恶意软件防护For more information, see the Azure Security Benchmark: Malware defense.

8.1:使用集中管理的反恶意软件8.1: Use centrally managed antimalware software

指导:Microsoft 反恶意软件是在为 Azure 服务(例如 Azure 机器学习)提供支持的基础主机上启用的,但它不会针对客户内容运行。Guidance: Microsoft anti-malware is enabled on the underlying host that supports Azure services (for example, Azure Machine Learning), however, it does not run on customer content.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的计算资源,请使用适用于 Azure 的 Microsoft Antimalware 持续监视和保护你的资源。For compute resources that are owned by your organization, use Microsoft Antimalware for Azure to continuously monitor and defend your resources. 对于 Linux,请使用第三方反恶意软件解决方案。For Linux, use third-party antimalware solution. 另外,请使用 Azure 安全中心的数据服务威胁检测来检测上传到存储帐户的恶意软件。Also, use Azure Security Center's threat detection for data services to detect malware uploaded to storage accounts.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

8.2:预先扫描要上传到非计算 Azure 资源的文件8.2: Pre-scan files to be uploaded to non-compute Azure resources

指导:Microsoft 反恶意软件是在为 Azure 服务(例如 Azure 机器学习)提供支持的基础主机上启用的,但它不会针对客户内容运行。Guidance: Microsoft Anti-malware is enabled on the underlying host that supports Azure services (for example, Azure Machine Learning), however it does not run on customer content.

你需要负责预先扫描要上传到非计算 Azure 资源的任何内容。It is your responsibility to pre-scan any content being uploaded to non-compute Azure resources. Microsoft 无法访问客户数据,因此无法代表你对客户内容执行反恶意软件扫描。Microsoft cannot access customer data, and therefore cannot conduct anti-malware scans of customer content on your behalf.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

步骤 8.3:确保反恶意软件和签名已更新8.3: Ensure antimalware software and signatures are updated

指导:Microsoft 反恶意软件是针对为 Azure 服务(例如 Azure 机器学习)提供支持的基础主机启用并维护的,但它不会针对客户内容运行。Guidance: Microsoft anti-malware is enabled and maintained for the underlying host that supports Azure services (for example, Azure Machine learning), however, it does not run on customer content.

Azure 机器学习为各种计算资源甚至为你自己的计算资源提供不同的支持。Azure Machine Learning has varying support across different compute resources and even your own compute resources. 对于你的组织拥有的任何计算资源,请按照 Azure 安全中心的“计算和应用”内的建议进行操作,以确保所有终结点都具有最新的签名。For any compute resources that are owned by your organization, follow recommendations in Azure Security Center, Compute & Apps to ensure all endpoints are up to date with the latest signatures. 对于 Linux,请使用第三方反恶意软件解决方案。For Linux, use third-party antimalware solution.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

数据恢复Data recovery

有关详细信息,请参阅 Azure 安全基线: 数据恢复For more information, see the Azure Security Benchmark: Data recovery.

9.1:确保定期执行自动备份9.1: Ensure regular automated back ups

指导:机器学习服务中的数据恢复管理是通过已连接的数据存储上的数据管理进行的。Guidance: Data recovery management in Machine Learning service is through data managements on connected data stores. 请确保遵循有关已连接存储的数据恢复准则,以根据客户组织策略备份数据。Ensure to follow up data recovery guidelines for connected stores to back up data per customer organization policies.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

9.2:执行完整系统备份,并备份客户管理的所有密钥9.2: Perform complete system backups and backup any customer-managed keys

指导:机器学习服务中的数据备份是通过已连接的数据存储上的数据管理进行的。Guidance: Data backup in Machine Learning service is through data managements on connected data stores. 请为 VM 启用 Azure 备份,并配置所需的频率和保留期。Enable Azure Backup for VMs and configure the desired frequency and retention periods. 在 Azure Key Vault 中备份客户管理的密钥。Back up customer-managed keys in Azure Key Vault.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

9.3:验证所有备份,包括客户管理的密钥9.3: Validate all backups including customer-managed keys

指导:机器学习服务中的数据备份验证是通过已连接的数据存储上的数据管理进行的。Guidance: Data backup validation in Machine Learning service is through data managements on connected data stores. 请定期在 Azure 备份中执行内容数据还原。Periodically perform data restoration of content in Azure Backup. 请确保可以还原已备份的客户管理的密钥。Ensure that you can restore backed-up customer-managed keys.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

9.4:确保保护备份和客户管理的密钥9.4: Ensure protection of backups and customer-managed keys

指导:对于本地备份,请使用备份到 Azure 时提供的密码提供静态加密。Guidance: For on-premises backup, encryption at rest is provided using the passphrase you provide when backing up to Azure. 使用基于角色的访问控制来保护备份和客户管理的密钥。Use role-based access control to protect backups and customer-managed keys.

在 Key Vault 中启用软删除和清除保护,以防止意外删除或恶意删除密钥。Enable soft delete and purge protection in Key Vault to protect keys against accidental or malicious deletion. 如果将 Azure 存储用于存储备份,请启用软删除以在 blob 或 blob 快照被删除时保存和恢复数据。If Azure Storage is used to store backups, enable soft delete to save and recover your data when blobs or blob snapshots are deleted.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

事件响应Incident response

有关详细信息,请参阅 Azure 安全基线: 事件响应For more information, see the Azure Security Benchmark: Incident response.

10.1:创建事件响应指导10.1: Create an incident response guide

指导:为组织制定事件响应指南。Guidance: Develop an incident response guide for your organization. 确保在书面的事件响应计划中定义人员职责,以及事件处理和管理从检测到事件后审查的各个阶段。Ensure there are written incident response plans that define all the roles of personnel as well as the phases of incident handling and management from detection to post-incident review.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

10.2:创建事件评分和优先级设定过程10.2: Create an incident scoring and prioritization procedure

指导:Azure 安全中心为每条警报分配严重性,方便你根据优先级来确定应该最先调查的警报。Guidance: Azure Security Center assigns a severity to each alert to help you prioritize which alerts should be investigated first. 严重性取决于安全中心在发出警报时所依据的检测结果或分析结果的置信度,以及导致发出警报的活动的恶意企图的置信度。The severity is based on how confident Security Center is in the finding or the analytically used to issue the alert as well as the confidence level that there was malicious intent behind the activity that led to the alert.

此外,使用标记来标记订阅,并创建命名系统来对 Azure 资源进行标识和分类,特别是处理敏感数据的资源。Additionally, mark subscriptions using tags and create a naming system to identify and categorize Azure resources, especially those processing sensitive data. 你的责任是根据发生事件的 Azure 资源和环境的关键性确定修正警报的优先级。It's your responsibility to prioritize the remediation of alerts based on the criticality of the Azure resources and environment where the incident occurred.

Azure 安全中心监视:是Azure Security Center monitoring: Yes

责任:客户Responsibility: Customer

10.3:测试安全响应过程10.3: Test security response procedures

指导:定期执行演练来测试系统的事件响应功能,以帮助保护 Azure 资源。Guidance: Conduct exercises to test your systems' incident response capabilities on a regular cadence to help protect your Azure resources. 查明弱点和差距,并根据需要修改你的响应计划。Identify weak points and gaps and then revise your response plan as needed.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

10.4:提供安全事件联系人详细信息,并针对安全事件配置警报通知10.4: Provide security incident contact details and configure alert notifications for security incidents

指导:如果 Microsoft 安全响应中心 (MSRC) 发现数据被某方非法访问或未经授权访问,Microsoft 会使用安全事件联系信息联系用户。Guidance: Security incident contact information will be used by Microsoft to contact you if the Microsoft Security Response Center (MSRC) discovers that your data has been accessed by an unlawful or unauthorized party. 事后审查事件,确保问题得到解决。Review incidents after the fact to ensure that issues are resolved.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

10.5:将安全警报整合到事件响应系统中10.5: Incorporate security alerts into your incident response system

指导:使用连续导出功能导出 Azure 安全中心警报和建议,以便确定 Azure 资源的风险。Guidance: Export your Azure Security Center alerts and recommendations using the continuous export feature to help identify risks to Azure resources. 使用连续导出可以手动导出或者持续导出警报和建议。Continuous export allows you to export alerts and recommendations either manually or in an ongoing, continuous fashion.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

10.6:自动响应安全警报10.6: Automate the response to security alerts

指导:使用 Azure 安全中心的工作流自动化功能,针对安全警报和建议自动触发响应,以保护 Azure 资源。Guidance: Use workflow automation feature Azure Security Center to automatically trigger responses to security alerts and recommendations to protect your Azure resources.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:客户Responsibility: Customer

渗透测试和红队练习Penetration tests and red team exercises

有关详细信息,请参阅 Azure 安全基线: 渗透测试和红队演练For more information, see the Azure Security Benchmark: Penetration tests and red team exercises.

11.1:定期对 Azure 资源执行渗透测试,确保修正所有发现的关键安全问题11.1: Conduct regular penetration testing of your Azure resources and ensure remediation of all critical security findings

指导:请遵循 Microsoft 云渗透测试互动规则,确保你的渗透测试不违反 Microsoft 政策。Guidance: Follow the Microsoft Cloud Penetration Testing Rules of Engagement to ensure your penetration tests are not in violation of Microsoft policies. 使用 Microsoft 红队演练策略和执行,以及针对 Microsoft 托管云基础结构、服务和应用程序执行现场渗透测试。Use Microsoft's strategy and execution of Red Teaming and live site penetration testing against Microsoft-managed cloud infrastructure, services, and applications.

Azure 安全中心监视:不适用Azure Security Center monitoring: Not Applicable

责任:共享Responsibility: Shared

后续步骤Next steps