Zero Trust for a World of Machines: APIs, AI Agents, and Autonomous Action

AI agents are already in production, and many security teams are unaware of them. Developers are shipping agents that call APIs, invoke tools, and take autonomous actions faster than security can instrument, govern, or even inventory them. The blast radius has extended beyond what traditional solutions can see and secure. An unidentified, over-privileged agent can exfiltrate data, corrupt systems, or be weaponized via prompt injection — all through the same API calls your legitimate workloads make. Every new tool connection, OAuth grant, MCP server, and agent-to-agent connection expands the attack surface. While Zero Trust originated in a human-centered world, its fundamentals continue to be relevant and necessary for the agentic layer: verify every agent identity, enforce fine-grained authorization on every API call, enforce least-privilege access, and detect anomalous behavior before it becomes a breach.The good news is this framework can be implemented today using open standards and production-grade open source infrastructure. Join Numberline Security and Ory for a discussion on securing agentic AI systems. We'll cover: - Where agent deployments break such as identity gaps, over-permissioned APIs, and unsecured tool calls, with real attack patterns - Enforcing Zero Trust on agents including workload identity, scoped API access, fine-grained authorization policies, and runtime controls - A deployable reference architecture that encompasses agent identity provisioning, API security gateway, and centralized authorization, built with production-ready components from Ory Built for security architects, platform engineers, and AI strategy leads who need to move from “we should secure this” to “it’s secured.” You’ll leave with a threat model, a reference architecture, and actionable next steps.