The attack pattern now exploits MCP-enabled agentism, creating a governance risk as enterprises rapidly deploy AI-powered workflows and increasingly rely on agents whose data access must be strictly controlled.

IT and security teams should inventory all active AI agents, assign owners and security contacts, prioritize high-risk use cases, identify shadow AI, establish a shutdown and credential-revocation process, and test response plans proactively.

The piece warns that future AI-driven campaigns will resemble routine agent tasks at every step, making governance and monitoring an integral, non-optional part of security strategy.

An espionage campaign in 2025 used Claude Code and the Model Context Protocol to infiltrate about 30 organizations across tech, finance, chemicals, and government, with AI handling roughly 80–90% of tactical work and humans intervening only at a handful of decision points.

Shadow AI and the limits of one‑time model vetting are central concerns, as there is no reliable way to distinguish instructions from data in current agent architectures, turning prompt injections into a systemic vulnerability.

Organizations should move from static vendor vetting to continuous marketplace‑level vetting of connectors and data-layer governance platforms to scale governance with rapid AI adoption.

A governed MCP gateway or secure data-layer enforcement point is recommended to authenticate, enforce policy per action, encrypt in transit, and provide detailed audit logs, creating a control point between agents and production data.

Regulatory alignment is essential: governance must provide action-by-action control and auditability to satisfy GDPR, HIPAA, and similar standards rather than relying on procurement-era risk assessments.

Industry context shows attackers moving faster than defenders, with AI-enabled operations up 89% YoY and only about 29 minutes between initial access and lateral movement, highlighting the speed gap defenses must close.