Reference List & Traceability of AI Assistant Security Incidents
一、运行时失控 / 自治失控
| 序号 | 标题 | 日期 | 链接 | 适用风险标签 |
|---|---|---|---|---|
| 1 | M365 Copilot 信息泄露漏洞(CVE-2026-42827) | 2026-05-21 | https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-42827 | 运行时失控 / 数据外泄 |
| 2 | M365 Copilot AI提示词注入漏洞(CVE-2026-26133) | 2026-03 | https://avd.aliyun.com/detail?id=AVD-2026-26133 | 运行时失控 |
| 3 | AI Red Teaming 代理程式(Microsoft Azure 官方文档) | 2026-04-22 | https://learn.microsoft.com/zh-hk/azure/ai-foundry/concepts/ai-red-teaming-agent?view=foundry-classic | 运行时失控 |
| 4 | Multi-Agent Offensive Attacks Go Live: How Collaborative Rogue AI Agents Are Breaking Enterprise Defenses | 2026-04-05 | https://agentmarketcap.ai/blog/2026/04/05/multi-agent-offensive-attacks-collaborative-rogue-agents | 运行时失控 / 数据外泄 |
| 5 | 国家网络安全通报中心发布OpenClaw安全风险预警 | 2026-03-13 | https://news.cnr.cn/kuaixun/20260313/t20260313_527550897.shtml | 运行时失控 / 供应链投毒 |
二、数据外泄 / 数据泄露
| 序号 | 标题 | 日期 | 链接 | 适用风险标签 |
|---|---|---|---|---|
| 1 | M365 Copilot 信息泄露漏洞(CVE-2026-42827) | 2026-05-21 | https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-42827 | 运行时失控 / 数据外泄 |
| 2 | Statement on Strengthening Safeguards Within the Lilli Tool(麦肯锡官方声明) | 2026-03-11 | https://www.mckinsey.com/about-us/media/statement-on-strengthening-safeguards-within-the-lilli-tool | 数据外泄 |
| 3 | Shadow Escape: The First Zero Click Agentic Attack using MCP | 2025-10-22 | https://www.operant.ai/art-kubed/shadow-escape | 数据外泄 |
| 4 | 用户匿名对话AI 仍被扒出真实身份(南都测评) | 2026-04-07 | https://www.toutiao.com/article/7625786144077775375 | 数据外泄 |
| 5 | 国家网络安全通报中心发布OpenClaw安全风险预警 | 2026-03-13 | https://news.cnr.cn/kuaixun/20260313/t20260313_527550897.shtml | 运行时失控 / 供应链投毒 |
三、身份盲区 / 身份与权限失控
| 序号 | 标题 | 日期 | 链接 | 适用风险标签 |
|---|---|---|---|---|
| 1 | M365 Copilot 信息泄露漏洞(CVE-2026-42827) | 2026-05-21 | https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-42827 | 运行时失控 / 数据外泄 |
| 2 | Meta AI 客服漏洞遭黑客利用,超 2 万 Instagram 账号受影响 | 2026-06-10 | https://finance.sina.com.cn/tech/digi/2026-06-10/doc-iniaxeit4500399.shtml | 身份盲区 |
| 3 | 駭客靠「一句話」就接管 Instagram:Meta AI 客服機器人如何成為 2 萬個帳號的後門 | 2026-06-07 | https://www.studioglobal.ai/zh-Hant/discover/answers/what-two-separate-security-flaws-affected-instagram-6a25d1345b21f12644550b8b | 身份盲区 |
| 4 | Multi-Agent Offensive Attacks Go Live: How Collaborative Rogue AI Agents Are Breaking Enterprise Defenses | 2026-04-05 | https://agentmarketcap.ai/blog/2026/04/05/multi-agent-offensive-attacks-collaborative-rogue-agents | 运行时失控 / 数据外泄 |
| 5 | 国家网络安全通报中心发布OpenClaw安全风险预警 | 2026-03-13 | https://news.cnr.cn/kuaixun/20260313/t20260313_527550897.shtml | 运行时失控 / 供应链投毒 |
四、供应链 / 插件投毒
| 序号 | 标题 | 日期 | 链接 | 适用风险标签 |
|---|---|---|---|---|
| 1 | Update on Mercor Security Incident(官方通报) | 2026-06-27 | https://www.mercor.com/blog/update-on-mercor-security-incident/ | 供应链投毒 |
| 2 | CVE-2026-27966(Langflow CSV Agent 节点 RCE 漏洞) | 2026 | https://nvd.nist.gov/vuln/detail/CVE-2026-27966 | 供应链投毒 / 运行时失控 |
| 3 | 科技巨头AI代理存重大安全漏洞:可被劫持窃取数据、长期操控 | 2026-04-21 | https://www.toutiao.com/article/7631026174105354771 | 供应链投毒 / 身份盲区 |
| 4 | 科技巨头AI代理存重大安全漏洞:可被劫持窃取数据、长期操控 | 2026-04-21 | https://www.snowbeasts.com/blog/details/1686 | 供应链投毒 / 身份盲区 |
| 5 | 国家网络安全通报中心发布OpenClaw安全风险预警 | 2026-03-13 | https://news.cnr.cn/kuaixun/20260313/t20260313_527550897.shtml | 运行时失控 / 供应链投毒 |
