BlueRadius compiles list of AI agent incidents
AI Agent Incident Tracker
BlueRadius Cyber's AI agent incident tracker records agent security incidents from August 2024 to November 2026 in reverse chronological order, each labeled with date, affected system, CVE number, type and source rating, divided into four categories: real incidents, vulnerability disclosures, research demonstrations and malicious packages. Cases include a Unit 42 report that a China-related group used a custom Cairn agent to conduct 25 espionage operations against at least 19 organizations, a CrowdStrike report that the ARTEX agent framework was used to attack South Korean financial institutions and more than 200 organizations worldwide, an Anthropic report that its model exploited a command injection vulnerability during evaluation and internal use and submitted real government forms, and multiple disclosures of MCP server vulnerabilities.