The announcement describes a development in AI infrastructure as Teleport introduces two components in its Agentic Identity Framework: the LLM Proxy and Delegated Identity. These are presented as addressing challenges in AI agent deployments by incorporating identity, access control, and auditing at both the decision-making and permissions levels.
Historically, a large portion of AI development has focused on LLM gateways—systems that manage traffic, cost, and prompts in front of model providers. By contrast, oversight of agent behavior when interacting with production infrastructure has received less attention. The LLM Proxy is described as adding an enforcement layer connected to identity and zero trust systems that govern access to infrastructure such as databases and cloud APIs.
LLM Proxy: Control of Agent Actions
The LLM Proxy functions as an intermediary between an agent and its inference endpoint. It provides visibility and enforcement at the point where agent requests are made, checking each request and response and logging them into Teleport’s audit trail. With an allow-list mechanism, it can restrict which agents are permitted to access specific inference endpoints, and records are maintained for auditing purposes.
Delegated Identity: Scoped Access Control
Delegated Identity allows operators or systems to assign defined permissions that determine what an agent can access. Instead of broad or persistent credentials, an agent is given a specific delegated identity tied to particular tasks. Activity is logged and associated with the originating identity and task, supporting a least-privilege access model.
This approach is intended to reduce exposure if an agent is compromised, by enforcing scoped, time-bound access patterns aligned with zero trust and just-in-time (JIT) access principles.
The capabilities are available through the Beams public beta and are positioned as part of the Agentic Identity Framework for use in production-oriented environments.
