The January 2026 AI Security Incident Roundup released by PointGuard AI Research Labs shows a significant increase in the number and severity of AI security incidents this month, with several incidents scoring over 7.0 on the AI Security Severity Index (AISSI).Among them, Model Context Protocol (MCP)-related vulnerabilities emerged as a major threat category, including Clawdbot MCP exposure, JamInspector control plane flaws, and other high-risk incidents. In addition, the ServiceNow "BodySnatcher" vulnerability (CVE-2025-12420) and the Microsoft Copilot "Reprompt" attack received high risk ratings of 8.7 and 8.3 high-risk ratings, respectively. The report notes that attackers have shifted from experimental exploration to systematic exploitation of AI workflows, that cue injection attacks are still evolving, and that the majority of high-risk incidents stem from toolchain and protocol weaknesses rather than the models themselves.
