# 专题日跟踪 · 安全 · 2026-09-03

> 日期：2026-09-03（Asia/Shanghai）。原文摘录保持源语言。不改写数字。

## 摘要

- **CSA Top Threats 2026（2026-08-13）**：IAM 升至第 1；AI-Enhanced Attacks 新进第 2；AI System Compromise 第 6（prompt injection / 模型窃取）。基础设施类威胁跌出榜。
- **VMware Explore 2026（2026-08-31）**：AgentMinder 宣布 GA；vDefend 发现 MCP/LLM/工具与影子 AI；Avi 限制未授权工具。能力绑 Private AI Cloud / VCF。
- **GB/T 25068.7-2025**：2025-12-02 发布，**2026-07-01 实施**，网络虚拟化安全。
- **深信服 HCI 6.10.0**：智能微隔离（近 30 日访问模式）+ Windows vTPM 2.0。英文博客抓取被 Cloudflare 拦截，能力以公开摘要与支持站 Release Note 日期 2026-06-26 为准。
- **SmartX + 海光 HCT（2026-04-22）**：芯片内生国密 + 榫卯 6.3 直通设备 HA。不是 CSV/TEE。
- **Nutanix Flow CNI**：FNS 策略覆盖 Kubernetes；公开称含于 NCI Pro。
- **HPE VME + Aruba CX**：微隔离卸载到分布式服务交换机。
- **Scale AcuVigil**：边缘托管网络分段 / PCI 叙事，不是 HyperCore DFW。

## 原文摘录

### [Artificial Intelligence (AI) Emerges as an Attack Enabler and Target…](https://cloudsecurityalliance.org/press-releases/2026/08/13/ai-emerges-as-an-attack-enabler-and-target-in-csa-2026-top-threats-report)

- 来源：CSA Press Release / 2026-08-13

> The addition of two new AI-related issues — AI-Enhanced Attacks (#2), which examines the use of AI to improve or automate attacks, and AI System Compromise (#6), which addresses AI systems as assets that can be compromised, manipulated, or abused — underscores this concern.
>
> 1. Inadequate Identity and Access Management (Identity & Access Management in 2024 survey) (up from #2)
> 2. AI (Artificial Intelligence)-Enhanced Attacks (new)
> …
> 6. AI System Compromise (new)
>
> The rapid growth of non-human identities, machine-to-machine interactions, and autonomous decision-making has now surpassed what traditional management and oversight processes were designed to handle.

### [Identity, AI Lead CSA's 2026 Cloud Threat List](https://virtualizationreview.com/articles/2026/08/24/identity-ai-lead-csas-2026-cloud-threat-list.aspx)

- 来源：Virtualization Review / 2026-08-24

> The CSA Top Threats Working Group surveyed 507 security professionals and used their responses to rank 23 cloud security issues, producing a Top 11 led by identity, AI, third-party resources and APIs.
>
> CSA defines the issue as the exploitation, manipulation, degradation or unauthorized control of systems containing AI or machine-learning components, including models, prompts, training and inference data, connected tools, orchestration logic and decision pipelines. Attack vectors identified in the report include prompt injection, data or model poisoning, adversarial inputs, model theft and compromise of connected APIs, plugins or pipelines.

### [Broadcom Unveils AgentMinder…](https://www.globenewswire.com/news-release/2026/08/31/3353342/19933/en/broadcom-unveils-agentminder-an-enterprise-solution-for-ai-agent-governance-and-runtime-control.html)

- 来源：GlobeNewswire / LAS VEGAS, Aug. 31, 2026

> Broadcom Inc. … today introduced AgentMinder, a new solution that acts like a traffic controller for autonomous artificial intelligence (AI) agents. … AgentMinder independently verifies agent identity and authorizes each action against its declared mission, intent, context, and current risk before the action reaches an enterprise resource.

### [Broadcom Delivers End-to-End Security, Identity, and Observability for Agentic AI](https://www.globenewswire.com/news-release/2026/08/31/3353355/19933/en/broadcom-delivers-end-to-end-security-identity-and-observability-for-agentic-ai.html)

- 来源：GlobeNewswire / 2026-08-31

> VMware vDefend delivers Zero Trust lateral security for AI workloads running on VMware Cloud Foundation. The following new vDefend enhancements will extend its ZeroTrust lateral security to agentic AI workloads.

### [Discover New Sangfor HCI 6.10.0 Part 3…](https://www.sangfor.com/blog/cloud-and-infrastructure/discover-new-sangfor-hci-6100-part-3-raising-the-bar-for-hyperconverged-infrastructure-security)

- 来源：Sangfor 博客（WebFetch 2026-09-03 返回 Cloudflare 拦截页；下列依据检索摘要，规划引用须再核原文）

> The marquee security enhancement in this latest release is the introduction of smart micro-segmentation. … automatically generates access control policies based on the actual application access patterns observed over the past 30 days.
>
> One major enhancement is support for vTPM 2.0 on Windows operating systems.

### [SmartX&海光信息率先推出“芯片内生加密+虚拟化高可用”联合解决方案](https://www.smartx.com/blog/2026/04/smartx-hygeon-hci-ha-solution/)

- 来源：SmartX 博客 / 2026年4月22日

> 该方案可基于设备关联标识机制，对加密虚拟化设备进行统一识别、调度与重建，使相关业务虚拟机在获得芯片级数据加密能力的同时，也具备跨节点自动故障恢复能力。
>
> 基于海光信息 CPU 内置 HCT 密码协处理器，在芯片内完成 SM2/SM3/SM4 等国密算法运算，从而替代传统外置硬件加密机。

### [GB/T 25068.7-2025](https://ndls.cnis.ac.cn/standard/detail/bc8844380675298dae12ad80b01a3b6f)

- 来源：国家标准馆

> 发布日期：2025-12-02 实施日期：2026-07-01
> 中文名称：信息技术 安全技术 网络安全 第7部分：网络虚拟化安全

### [Enterprise-Grade Network Security Policy for Kubernetes: FNS Comes to Flow CNI](https://www.nutanix.com/tech-center/blog/enterprise-grade-network-security-policy-for-kubernetes-fns-comes-to-flow-cni)

- 来源：Nutanix Tech Center

> Flow CNI delivers full FNS policy parity for container workloads. AppType and Isolation policies now work on pods and services, in both monitor and enforce mode… Flow CNI is available now, included with NCI Pro at no additional cost.

## 对专题档案的影响

- 安全专题须把 **Agent/MCP 身份** 从「AI 附录」提升为与 DFW 并列的选型条件。
- SmartX HCT 稿不得写入机密计算「有」。
- 深信服 6.10 智能微隔离与 vTPM 写入对照表；英文页抓取失败记为异常。
- 国标 25068.7 实施日已过，规划话术可改「符合即将实施的国标」为「已实施」。