What happened
OpenAI announced (Sept 10, 2026) the Agents API public beta, a managed service exposing session orchestration, long-running sessions, sub-agent coordination, lazy tool loading, and crash recovery — the same harness underlying OpenAI's own Codex-style agents — with agent/environment/session/event primitives and MCP tool support.
Why it matters
As a managed agent-runtime service from a top-tier AI lab, it will materially expand the population of production agentic deployments that downstream AI-security vendors (guardrails, MCP gateways, agent-IAM) must cover; it is infrastructure, not itself a security product, so materiality is limited to watch-list status.
Applicability
AI-security vendors building agent guardrails/MCP proxies should validate compatibility with the Agents API's session/tool-call model now that it is in public beta.