Discussion

Linx Security adds controls for AI agents, including the ones nobody approved

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#14040

Linx Security has announced four capabilities for finding and governing AI agents across enterprise systems, including unsanctioned agents and controls on what agents can do. The practical shift is from tracking access to putting decisions about agent actions into the systems that enforce them.

Watch Desk analysis

What happened

The capabilities, described by Help Net Security in its 17 November report, include Shadow AI Agent Detection, which Linx says can find agents running outside sanctioned platforms, including on endpoints and in cloud services. Findings are added to its Identity Graph with an owner and information about potential exploitation and remediation.

Agent Runtime Control works with an organisation’s existing AI gateway, or Linx’s own MCP Gateway, to allow or block tool calls based on the agent’s identity, risk and intended action. Linx also introduced Autopilots, which teams can give a defined mission and scope, with choices about which actions run autonomously and which need human approval. Prompt-to-Workflow turns a plain-English description into a workflow that teams can inspect and dry-run before activation.

Linx says its customers already govern more than 25,000 AI agent identities in production and that governed identities grew 450 times in six months. Those figures and the product capabilities are company claims reported by Help Net Security.

Why it matters

AI agents can be created, changed and reactivated quickly, then act across company tools and data. That makes familiar access reviews and human-paced approval processes a poor fit on their own. The announced tools aim to give security teams a way to discover agents, judge their permissions and control actions without replacing the gateway an organisation already uses.

The useful details are the proposed controls, not the grand promise of making agents safe. A dry-run and configurable approval rules give teams checkpoints before an automated workflow goes live; they do not, by themselves, establish how well the system detects every agent or blocks every risky action.

Our read

This is a concrete enterprise security development because it addresses how AI agents are identified and governed, rather than simply adding another agent to the roster. The most compelling idea is to make action control part of identity governance. The test will be whether teams can see why a particular action was allowed or blocked, and whether that decision holds up in messy real-world systems.

What to watch

  • Whether Linx publishes technical detail on detection coverage and how its action decisions are enforced.
  • How the Autopilots’ approval settings and dry-runs work in customer deployments.
  • Whether customers or independent assessments provide evidence of detection accuracy and operational results.

Discussion spark: Should enterprise AI agents be allowed to take routine actions autonomously once they have a defined scope, or should human approval remain the default for every consequential step?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.