AI Agents Expand Cyberattack Surface Globally

Security researchers and industry leaders warned this week at a Cloud Security Alliance briefing that the rise of AI agents and Anthropic’s November 2024 Model Context Protocol (MCP) standard is broadening the cyberattack surface. They cited findings that 95% of MCP deployments run on unmanaged endpoints, honeypots logged over 91,000 probing sessions on LLM infrastructure, and nation-states like China and Iran are accelerating offensive AI capabilities.