Microsoft Treats AI Agents as Insider Threats in New Security Framework

Security Affairs reports multiple October 2026 cybersecurity developments, including active exploitation of a SiteGuard Server Edition command-injection flaw enabling root or SYSTEM execution, a local OpenBSD privilege-escalation exploit chain, credential theft targeting Microsoft 365 through ConnectWise identities, an unauthenticated Broadcom Wi-Fi vulnerability enabling router takeover, and CikaBika Bot campaigns using fake documents, Discord, and Ngrok for remote access and data theft. Additional coverage addresses AI-agent security, deepfake detection, and AI data-scraping policy changes.