What happened
WitnessAI launched Agentic Control on June 17 — a runtime enforcement layer that governs how AI agents interact with enterprise tools and MCP servers. Capabilities include: automated agent discovery across IDEs, apps, agent frameworks, and cloud-built custom agents; an MCP Catalog that scores known tools against OWASP and CVE risk classes; org-wide allow/block lists for MCP servers and tools; runtime inspection and enforcement of agentic conversations and tool invocations; and a tamper-evident audit trail. The product closes the gap between WitnessAI's earlier (January) discovery layer and actual policy enforcement — blocking calls before they land, not just naming the agent after the fact.
Why it matters
Most enterprise security tooling cannot parse MCP communications, tool invocations, or agent-to-agent workflows. As agentic AI deployments accelerate, compromised or misconfigured agents can inherit full user permissions and act at machine speed — making runtime enforcement at the tool boundary the critical missing control. WitnessAI is the first to ship a combined discovery + OWASP/CVE-scored catalog + runtime block capability in a single platform.
Applicability
CISOs and security architects deploying Claude Code, Cursor, or custom MCP-connected agents in enterprise environments should evaluate immediately; especially relevant for organizations under EU AI Act compliance timelines.