Teleport unveils identity framework for agentic AI security

Teleport has launched an agentic identity framework that sets out policies, practices, developer tools and a reference architecture for running autonomous and semi-autonomous AI agents as trusted identities across cloud and on-premises infrastructure.

The company said organisations face a growing security gap in always-on systems that invoke tools, access sensitive data, and delegate tasks without direct human involvement. Teleport positioned identity as the foundation for trust in these deployments.

Teleport based its announcement on a survey of more than 200 infrastructure leaders. It said 69% of respondents believe widespread AI adoption will require significant changes to identity management, while 2% disagreed. The company said it will publish full survey results later.

Identity shift

Agentic AI introduces a different access pattern compared with traditional workloads, according to Teleport. The company said agents operate across environments at scale and interact with tools and data in ways that existing identity and access controls did not anticipate.

It said organisations deploying agents often rely on static secrets, hard-coded credentials, or bespoke configurations layered on top of legacy identity and access management and privileged access management tools. Teleport linked that approach to identity fragmentation, secrets sprawl and reduced visibility as deployments grow.

Teleport said its approach treats AI agents as "first-class identities" and uses cryptographic controls and ephemeral credentials. It also highlighted auditability and authentication. The company said the framework aims to reduce the risk of data compromise, misuse and external adversarial threats in production environments.

Teleport also positioned its infrastructure identity platform as an underlying layer for the framework. The company said it uses a hardware root of trust and replaces static credentials with ephemeral identities. It described that model as "zero trust authentication" and "zero standing privileges", with real-time visibility into identity behaviour across infrastructure.

"A unified identity layer is a prerequisite to deploying AI within enterprise infrastructure environments," said Ev Kontsevoy, Co-Founder and CEO, Teleport. "Deploying AI on top of fragmented credentials and identity silos is a recipe for secrets and data leakage."

Standards focus

Teleport said the framework adopts open standards, including MCP and SPIFFE. It linked that decision to reducing vendor lock-in and maintaining compatibility as the ecosystem evolves.

The company also described a "governed MCP and LLM control plane" as part of the model. It said this layer unifies identity governance across agents, tools and data. Teleport associated it with budgets, rate limits and guardrails.

The framework also addresses operations at scale, according to Teleport. The company said it targets consistent security and reliability in production deployments across cloud and on-premises environments.

Security debate

Teleport drew a distinction between its identity-led approach and vendor strategies that focus on model safety, runtime monitoring, or post-incident risk detection. The company argued that those measures do not address how trust is established, enforced, and governed for agents within the production infrastructure.

One analyst said identity sits at the centre of agent security. "As organisations deploy autonomous AI agents, identity - not monitoring - becomes the primary security control. Without a unified identity foundation, agentic systems introduce unmanageable risk across data, infrastructure, and compliance," said Frank Dickson.

Teleport said the framework provides a roadmap for cryptographic identity, ephemeral privileges, access guardrails, auditability and real-time enforcement. It presented the framework as a guide for technology leaders planning production deployments.

The company said it expects the framework to evolve alongside industry and community needs. It also stated that it has published a review of the current and planned components of the Teleport Agentic Identity Framework.