AI agent runtime governance, explained properly.

What it means to govern AI agents at runtime: identity, least privilege, policy enforcement, and audit trails for software that acts on its own.

• What is AI Agent Runtime Governance?AI agent runtime governance is the control layer that decides what an AI agent is allowed to do at the moment it acts: identity per agent, policy enforced on every action before it executes, and an audit trail of every decision.
• AI Agent Least PrivilegeLeast privilege for AI agents: every agent starts with zero permissions and receives explicit, scoped grants — per agent, per system, per condition. Why traditional least privilege breaks for agents and how to apply it.
• AI Agent Audit TrailsAn AI agent audit trail is a complete, attributable record of every action an agent attempted: which agent, which tool, what parameters, when, what the control decided, and why. What auditors expect and how to build it.

The Model Context Protocol is how agents reach real systems. Threat models, authentication vs authorization, and securing MCP servers and tool calls.

• What is MCP Security?MCP security is the practice of securing the Model Context Protocol layer — the servers, clients, and tool calls through which AI agents reach real systems. The attack surface, the controls, and where enforcement belongs.
• The MCP Threat ModelA systematic threat model for the Model Context Protocol: trust boundaries between agent, client, server, and downstream systems, the attack vectors that cross them, and the mitigations that hold at each boundary.
• MCP Authentication vs AuthorizationMCP authentication proves who is calling; authorization decides what that caller may do on each tool call. The spec standardizes the first and leaves most of the second to you — conflating them is how agents get over-permissioned.

Claude Code, Cursor, and custom coding agents with production access: what can go wrong and how to contain it.

• Securing Claude Code with Production AccessSecuring Claude Code means constraining what the agent can execute — shell commands, file edits, MCP tool calls — rather than trusting model behavior. The attack surface, the layered controls, and a working policy.

Working examples — policy enforcement, tool-call logging, and least privilege on real MCP servers, using the SentnelOps SDKs.

• Add Policy Enforcement to an MCP ServerA working guide to enforcing policy on MCP tool calls: in-server middleware vs a proxy in front, a default-deny YAML policy, decision logging, and testing policy changes in CI before they guard production.