Microsoft's 2026 Digital Defense Report covers July 2025 to June 2026 and states that attackers are gaining AI benefits ahead of defenders. The median time from vulnerability discovery in the wild to weaponization has dropped to under 24 hours, and the number of CVEs in 2026 is expected to reach a record 72,000, while patching speed cannot keep up; Microsoft expects known unpatched vulnerabilities to accumulate for years. Phishing's share of intrusions investigated by Microsoft's incident response team rose from 7% a year ago to 23%, and exploitation of vulnerabilities in public-facing applications rose from 15% to 24%. The report also mentions s1ngulari. Subsequent information shows that the report's PDF document has been made public.
Starting October 1, 2026, Google stopped accepting new product vulnerability reports through the Open Source Software Vulnerability Rewards Program (OSS VRP), citing a significant increase in automated submissions, the vast majority of which are invalid and mainly AI-generated, overwhelming review engineers and open source maintainers. The program covers the latest versions of public repositories and repository configurations under Google's GitHub organizations, including projects such as Go, Angular, and Protocol Buffers. Google explained the reason for the pause in an X post and plans to announce subsequent adjustments in the first quarter of 2027. The rules page has removed product vulnerability reward amounts of $500 to $7,500 for flagship projects and $101 to $3,133.7 for important projects. Supply chain poisoning reports are still accepted; vulnerabilities involving Google Cloud products can be submitted to Cloud VRP instead; reports submitted before October 1 are unaffected. Google did not provide the proportion of AI-generated reports.
HHiddenLayer ResearchHiddenLayer Research · Attention1212
HiddenLayer embeds AI security capabilities directly into teams' existing workflows. Its AI Supply Chain module scans models connected in the Azure AI Foundry catalog, checking for tampering, backdoors, and known vulnerabilities; on AWS it covers Bedrock, Bedrock Agents, SageMaker, and the Strands framework; in Databricks it automatically scans models against Unity Catalog. Detection results can be exported to Splunk and Microsoft Sentinel.
安全人工智能联盟（CoSAI）成立两年之际公布其智能体安全工作进展，涵盖《The Future of Agentic Security: From Chatbots to Autonomous Swarms》报告及在 RSAC 2026 发布的 MCP 安全和智能体身份研究成果。该组织通过四个工作流推进最佳实践共享，并新增 OWASP、AI Alliance 和云安全联盟加入技术指导委员会。下一步将发布面向 AI/ML 供应链信任与溯源的工件完整性成熟度模型、Zero Trust for AI Systems 以及 MCP 安全指南第二版。
英国 NCSC 联合国际伙伴发布《谨慎采用智能体 AI 服务》指南，建议组织从小规模、低风险任务起步，并从一开始就套用既有网络安全控制。指南指出智能体 AI 能访问数据源、记忆上下文、调用工具并自主行动，甚至创建子智能体，因此继承了 LLM 的越狱和提示注入风险，同时因自主性和复杂度提高而扩大攻击面、更难预测、测试和治理。新增风险包括更广的系统与数据访问权限、目标被意外解读导致的行为不可预测、行动速度快于人工审查、以及行为难以解释。具体措施包括最小权限、限制访问范围与可执行动作、避免长期凭证、使用安全默认配置、管理第三方组件与模型的供应链风险、监控异常行为、对部署做威胁建模并把智能体失控纳入事件响应计划。指南强调人类仍需对部署决定、授予的权限、防护措施和运行后果负责，并明确谁拥有、谁批准访问、谁监控、谁审查事件以及谁有权停止智能体。
The European Union Agency for Cybersecurity (ENISA) draft 'AI-assisted software development technical advisory' version 0.4 is dated September 2026 and proposes embedding secure-by-design expectations into AI-assisted development workflows for coding assistants and agents. The document outlines components such as models, coding assistants and agents, tool integration (including MCP), and context and project instructions, and organizes adversarial threats under STRIDE—spoofing, tampering, information disclosure, and elevation of privilege—as well as non-adversarial risks such as functionality available but insecure, incomplete context, hallucination dependence, and insufficient review. Its core method has four steps: identify the lifecycle process and security requirements corresponding to the activity; communicate these requirements to the assistant via prompts or reusable skills; verify outputs through review and approval with checks such as SAST and dependency scanning; and record evidence. The annex uses dependency selection as an example, showing how to write package manager security recommendations as SKILL.md skill content, and notes that skills do not guarantee secure outputs and require ongoing maintenance.
CNA, Monetary Authority of Singapore · 3 sourcesCNA · 3 sources · 5 reports · 
