Obsidian Launches Runtime Policy Control for AI Agents
Obsidian Security's new engine lets enterprises centrally allow, block, or require human approval for AI agent actions.
Updated on Nov. 10, 2026 in Artificial Intelligence

Obsidian Security announced its Runtime Policy Decision Engine, which evaluates AI agent actions against centrally managed policies. The company also expanded runtime coverage to OpenAI Codex and added its MCP Security Proxy and Policy Decision API.
Why it matters
Enterprises face risk from employees using platforms to write code and launch workflows, and from third-party vendors embedding agents into their products. Centralized policy control aims to give organizations a single point of oversight over those actions.
The engine uses identity, employment, application context, and customer-provided data classification labels to decide whether an agent action is allowed, blocked, or sent for human approval where supported. Independent verification of enforcement accuracy has not been reported.
The players
Obsidian Security
Obsidian Security is a Palo Alto, California-based cybersecurity company that says it is trusted by Fortune 1000 and Global 2000 enterprises.
Cloudflare
Cloudflare is an internet infrastructure and security company that is partnering with Obsidian to provide policies for agentic access through the Obsidian MCP Proxy.
The details
The MCP Security Proxy checks agent tool calls before they reach business applications, while the Policy Decision API extends Obsidian's decisions to customer-operated gateways and custom agent tools. Obsidian is also partnering with Cloudflare to provide policies for agentic access through the Obsidian MCP Proxy.
Timeline
November 10, 2026: Obsidian announces the Runtime Policy Decision Engine.
The Tech Race
Obsidian's launch places it among security vendors racing to govern AI agents as tools like Model Context Protocol let agents call enterprise applications directly. The announcement extends that governance to agents built into major enterprise platforms such as Salesforce, Microsoft 365, and ServiceNow.
Companies using AI agents may gain a way to approve, block, or review agent actions before they reach business systems like Microsoft 365 or Salesforce. For employees, that could mean some agent-driven tasks require a human sign-off before completion.
The takeaway
Organizations deploying AI agents should consider whether they have a central point of control over what those agents are permitted to do. Reviewing which tools and platforms agents can access is a practical first step.
Further reading
Explore more coverage in Artificial Intelligence for broader context on how agent governance is evolving.









