Greta Wexler

Proper governance infrastructure speeds up agent deployment, not slows it down.

Governance, not protocol, is why MCP pilots stall before production.

Trust in AI systems requires deliberate design, not just technical competence.

Organizations must inventory all deployed agents before any other security controls can work.

Frameworks aren't bureaucracy anymore; they're the operational core every board now requires.

Early interception stops tool-poisoning attacks; late filtering just documents the damage.

Attackers hide malicious instructions in tool metadata that AI agents trust automatically.

How MCP collapses the N×M integration problem into N+M through standardized roles.

NIST 600-1 wasn't built for AI systems that act on the world—here's where the framework breaks.

Most MCP servers skip identity verification entirely, leaving authorization meaningless.

Agents reach production faster than security can build guardrails around them.

MCP standardizes tool connections differently for generative and agentic AI workloads.

Exposure of thousands of unsecured MCP servers signals urgent need for layered risk assessment.

Most organizations treat AI agents as privileged identities without governing them like it.

Most MCP servers ship with dangerous defaults that leave them wide open to attack.