Teleport Beams, a Trusted Runtime Securing Agents in Production Infrastructure, Now Generally Available

Unlike traditional sandboxes that focus on VM level isolation of agentic workloads and credential management, a trusted runtime gives agents delegated identity and controls access and behavior at the swarm level without leveraging secrets.

OAKLAND, CA - November 10, 2026 - Teleport today announced the general availability of Teleport Beams, a trusted runtime for securing AI agents in production infrastructure. Beams has been in public beta since March 2026.

“Most enterprises have not connected agents to their production infrastructure. They are blocked by a lack of delegated identity, strong access control without secrets and with full audit and traceability of intent to action. When the cost of an error is high, traditional sandboxes don’t cut it. Beams brings these missing capabilities to the market,” said Ev Kontsevoy, CEO and co-founder of Teleport.

Four layers of control

As a trusted runtime, Beams protects agentic operations in production environments through four layers of control, that govern both individual and swarm behavior:

• Control Identity: Each runtime has a built-in delegated identity from a user or another agent. Agents start with zero standing privileges and propose execution plans and access requests. A human or an LLM judge reviews and revises each plan, and a separate runtime executes it with only the approved roles.
• Control Access: Beams connects agents to cloud accounts, Kubernetes clusters, PostgreSQL databases, MCP servers, and more, with no secrets or passwords in the agent’s environment. It discovers available resources automatically and authenticates through IAM, Kerberos, X.509, and other identity-based methods. The LLM gateway handles model selection and captures LLM and protocol-level traces of everything the agent does.
• Control the Swarm: Beams detects goal hijacking, emergent coordination between agents, and other harmful behavior, providing security teams with a readable account of what agents did and whether it matched what they were meant to do. It captures all messages sent to inference models, including LLM traces and protocol-level messages, and scores each Beam session against the MITRE ATT&CK framework to generate risk summaries. This scoring uses both built-in and customizable classifiers that define rules that flag behavior that departs from an agent’s objective. 
• Control the Environment: Agents run in ephemeral micro-VMs with virtualized networking that controls egress, ingress, DNS, and captures all protocol traffic in the audit record. VMs expire, base images are rebuilt regularly to reduce CVEs, and Beam Phases let teams restart from checkpoints or fork runtimes for parallel and recursive workflows, with the same isolation applied to every branch.

Differentiated Control

Beams innovative features address some of the key questions faced by AI practitioners: 

• Natural-language classifiers: Agentic Classifiers define rules that flag behavior that departs from an agent’s objective. Beams includes seven pre-built rules, such as “secrets exposed inline,” “destructive actions,” and “privilege escalation,” and also supports customer-defined rules.   
• Agent state & memory: Phases enables work to continue across beams without starting from scratch, by taking a snapshot of the artifacts and memory of an expiring beam and using it to set up the next one. Phases can also be used to explore diverging ideas safely through forking and parallel work. 
• Kill switch: Teams can stop everything at once: terminate the agent’s activity, its actions and all connections to any production infrastructure. 
• Execution plans: Rather than inject human-in-the-loop review for every agent step, agents present execution plans for approval, and then generate new action requests for desired steps outside of the approved plan.

"To scale agents in production, enterprises need controls that hold at the moment of action: model choice with a record of which model handled each invocation, preserved state and memory so a stop can be reversed, a kill switch that halts an agent mid-execution, and approval tiered by risk so humans are able to review high-impact actions. Platforms that build these into the runtime are helping move agentic AI from prototype to production." said Grace Trinidad, Research Director for AI Security & Trust at IDC.

Limiting the blast radius

Beams doesn’t try to guarantee what a model decides. It limits what an agent can reach, shows what it did, flags behavior that drifts from its objective, and ends its authority when the task is complete. A mistake or a hijacked agent has a small, visible blast radius. That is what lets security teams approve agents and move them from prototype to production.

Built on a proven foundation

Beams is built on the Teleport Infrastructure Identity Platform, in production at more than 700 enterprises including Nasdaq, DoorDash, and Elastic, extending the principles of zero trust to how agents behave. Teleport’s zero trust foundation establishes a unified identity layer for humans, machines, workloads, and agents, eliminating static credentials and standing privileges from infrastructure environments.

Availability

Teleport Beams is generally available as of November 9, 2026, at https://www.beams.run.

About Teleport

Teleport, the AI Infrastructure Identity Company, prepares organizations for an AI future by establishing a unified identity layer for infrastructure, with humans, machines, workloads, and AI agents secured cryptographically with a hardware root of trust rather than vulnerable credentials. By replacing fragmented identity and access management systems with Infrastructure Identity, Teleport scales zero trust across cloud and on-prem environments, eliminating the complexity and risk created by identity fragmentation and credential sprawl. Teleport protects infrastructure from identity attacks, accelerates engineering by reducing infrastructure complexity, and secures non-deterministic agentic workflows. Headquartered in Oakland, CA, Teleport operates globally, with industry-leading customers such as Nasdaq, Carta, and Elastic. For more information, visit goteleport.com or follow @goteleport.