JFrog AI Security Now Scans Laptops to Catch Shadow AI Agents Bypassing Central Repo Controls

JFrog announced an expansion of its Shadow AI detection capabilities to include developer workstations, addressing a gap where unvetted artificial intelligence assets previously operated without central oversight. This move targets locally stored files that often escape the scrutiny applied to code in official repositories, aiming to bring visibility to the autonomous actions of coding agents on individual machines.

The risk

According to JFrog research, seemingly benign files such as ZIP archives, Model Context Protocol servers, and newly created agents frequently circulate between developers via chat tools or direct sharing. These items can contain AI skills that were never registered as approved artifacts within an organization’s applications. Because these assets reside on private workstations, they often lack the governance applied to central systems, creating an unmonitored pathway for potentially harmful code to execute during development tasks.

Who is exposed

Software development teams that rely on AI coding agents face the primary exposure, as these tools can instantly introduce unvetted scripts and plugins from sources like GitHub or Slack threads. The risk extends to any environment where developers use locally minted agents or custom connectors to interact with external systems such as databases or project management tools. These local paths operate independently of the managed supply chain, leaving security teams without visibility into the specific capabilities or origins of the running code.

Cost and compliance

JFrog notes that assets such as Hooks allow individual authors to determine what runs on a machine every time an agent is invoked, often without consulting the broader team. This lack of consultation complicates compliance efforts, as a locally built connector to a service like Jira may violate organizational policies if it originates from an unapproved repository. The distinction between an acceptable, approved integration and a risky, self-made one becomes difficult to maintain without automated detection on local endpoints.

What should you do?

Managers should audit current policies to ensure they explicitly cover locally developed AI assets and scripts shared between team members. Security teams need to verify whether their current scanning tools can inspect files on developer laptops or if they are limited to central repositories. Organizations must define clear guidelines for what constitutes an approved AI skill or connector to prevent policy violations.

The background in plain language

Teams usually manage software components through a central hub where every file is scanned for vulnerabilities before use. However, AI tools allow developers to create and share small programs or instructions directly on their computers, bypassing that central check. This creates a blind spot where unapproved code can run without anyone noticing. To maintain security, organizations must extend their monitoring to individual workstations, ensuring that locally created tools meet the same safety standards as those from official sources.

Background: AI agents

AI agents are software systems that perceive their environment, make decisions, and execute actions to achieve specific goals. Unlike standard models, they maintain internal state and interact with external tools, creating unique security risks around tool execution, memory retention, and autonomous decision loops that require specialized controls.

Read the full guide: AI Agents Explained: Answers to Security and Architecture Questions

What to do and how to stay safe: JFrog

• Verify if your security scanning tools currently inspect files stored on developer laptops and local drives.
• Establish a policy that requires all AI skills and connectors to be registered in a central repository before use.
• Monitor for unauthorized execution of shell commands or scripts initiated by locally stored AI agents.
• Review sharing practices to ensure files transferred between colleagues are scanned for embedded AI assets.

Step-by-step guide: AI Governance Explained: Building Rules for Autonomous Systems

General security guidance from the Vector Tribune newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What new types of files is JFrog scanning for on developer machines?

JFrog is now scanning for ZIP files containing agent skills, locally developed MCP servers, and newly minted agents that reside on workstations.

How does Shadow AI differ from traditional third-party code risks?

Shadow AI assets are often introduced instantly and autonomously by developers or agents, bypassing the centralized build processes used for traditional code.

Can the new detection distinguish between different risk levels?

Yes, JFrog AI Security provides information on potential threats, such as distinguishing between a skill that reads files and one that executes shell commands.