Linx Security extends identity governance to AI agent actions

Help Net Security

Source published: Site published:

Original article

Share

Linx Security has announced four new features that bring every agent under identity control, whether it is managed or unmanaged and no matter which gateway it uses. Shadow AI Agent Detection finds agents that have not been approved. Agent Runtime Control decides whether each action an agent tries to take through an organization's existing gateway is allowed.

For identity teams, Autopilots for any use case let them put controlled agents to work. Linx says this is a first among major identity vendors. Prompt-to-Workflow turns a description written in plain English into a tested workflow that uses AI agents. These features expand identity governance beyond managing who can access systems to controlling what AI agents do across an organization.

Agent security starts with identity

Agents are created, given access and put to work very quickly. But security checks still follow human schedules, using support tickets, approvals and reviews that happen from time to time. Agents also do not fit the controls designed for people. They can exist only briefly, be created, renamed, changed or started again with little record of what happened. They do not follow a fixed script, so it can be hard to know what they will do next. They can also act on their own across company tools and data.

Many agents are examples of shadow AI. They run through personal accounts, software development pipelines and AI apps outside approved platforms. They may have no owner and receive no security review. Gartner predicts that more than 50% of enterprise AI projects will stop by 2028 because companies cannot solve problems with agent identities.

Instead of treating AI agents as a separate security problem, Linx extends its existing identity governance system to cover every human identity, non-human identity (NHI) and AI agent identity. Fixing this problem takes more than finding agents or watching what they do. Organizations need to understand each agent's identity, permissions, connections and actions in context. Linx uses this information in three steps.

First, build a graph: add every agent, whether approved or not, to one Identity Graph alongside all human and non-human identities. The graph also shows who owns each agent and whose access it uses. Second, make decisions using context: use the graph to assess risks and set rules for every tool call an agent makes and every change to its access. Third, control the action: apply those decisions through whichever gateway the agents use. Linx also gives identity teams their own agents and AI-powered workflows to do the work. Linx says its customers already govern more than 25,000 AI agent identities in live systems, and the number of governed agent identities grew 450 times in six months.

What's new

Shadow AI Agent Detection. Linx already keeps a list of agents on the platforms where they are built. Shadow AI Agent Detection also finds AI running outside those platforms, including on computers, in CI/CD pipelines, and in SaaS and cloud services. Each finding is added to the Linx Identity Graph. It includes the agent's owner, how an attacker might misuse it and how to fix the problem. This lets security teams manage unapproved agents just like any other identity.

Agent Runtime Control. Organizations can now use their own AI gateway. Linx connects to the gateway they already have and makes decisions about whether agent actions are allowed, using the open AuthZEN standard. Organizations that do not have a gateway can use the Linx MCP Gateway. Each time an agent tries to use a tool, Linx allows or blocks the request based on the identity behind the agent, the level of risk and what the agent is trying to do. Organizations do not need to replace their existing systems.

Autopilots for any use case. Identity teams can now add agents with specific jobs to their teams. They describe the agent's mission in everyday language, set limits on what it can do and choose which actions it can take by itself, which need human approval and which it must never take. Each Autopilot uses the information Linx has about the organization's identities and access. Before switching it on, teams can review its plan and test it without carrying out the actions.

Prompt-to-Workflow. Teams can describe a workflow in plain English. For example, they can ask for an agent to be turned off when its owner leaves the company. Linx builds the workflow step by step, shows the code behind it and tests it without running it before it goes live.

“AI agents are fundamentally changing the identity security problem,” said Israel Duanis, CEO of Linx Security. “Enterprises need to know not just who has access, but what every identity is connected to, what it's doing and why. That context is the foundation we've built at Linx, and we're now extending it to every AI agent.”

Share