JFrog Expands Shadow AI Detection to Developer Workstations and Local Assets

JFrog has expanded its AI security capabilities to include developer workstations, allowing organizations to detect shadow AI assets that reside on local machines rather than in centralized repositories. According to research published by JFrog, this update addresses the growing risk of unvetted AI components being used autonomously on developer laptops without central oversight or approval.

Previously, JFrog AI Security focused on detecting AI models within JFrog Artifactory repositories and monitoring calls to external AI services during build processes. This approach established a baseline for the managed path of the software supply chain. However, shadow AI often exists on private developer workstations that are not governed by central controls, creating a blind spot in traditional security monitoring.

The Rise of Local Shadow AI

Shadow AI refers to AI assets that have not been detected or registered as approved artifacts in an organization’s applications. These assets often include ZIP files shared between colleagues, locally developed Model Context Protocol (MCP) servers passed between developers, and newly minted agents stored locally. While these files appear harmless and are common in corporate software development, they can contain unapproved AI components that bypass standard security checks.

In the current AI era, developers can instantly introduce unvetted files, scripts, and plugins on demand. These assets often originate from GitHub, Slack threads, or are generated directly by the AI agent itself. This capability allows AI tools to circumvent the traditional, secure process where approved code and tools come from trusted sources and pass company security policies before use.

Risks of Ungated Executables

The introduction of unchecked assets creates significant security challenges because they operate outside the managed path. Skills, which are bundles of instructions defining how an agent performs tasks, and MCP servers, which connect agents to systems like databases or Slack, can be deployed without review. Plugins that bundle these components further complicate visibility.

JFrog notes that hooks represent a specific risk, as authors can unilaterally decide what runs on a machine every time an agent is invoked, without consulting the team. This creates a dichotomy between the managed path, where assets are scanned and vetted, and the local path, where shadow AI components go undetected. The new detection capability aims to bridge this gap by providing visibility into these local assets.

What should you do?

Security teams and platform administrators can now use JFrog AI Security to gain essential information about potential threats on developer machines. The tool provides details on the risk level of specific assets, such as whether a skill can run shell commands versus only reading files. It also checks organizational policy compliance, distinguishing between approved GitHub MCP servers and those built from random repositories, helping teams maintain control without hindering developer productivity.

What to do and how to stay safe: JFrog

• Audit local developer workstations for unapproved AI skills, MCP servers, and agents that bypass central repository controls.
• Implement tools that can detect and assess the risk of locally stored AI assets, such as those capable of executing shell commands.
• Review organizational policies to ensure that all AI connectors, including MCP servers, are vetted and approved before use on local machines.
• Monitor for the introduction of unvetted files and plugins from external sources like GitHub or Slack into the development environment.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

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

Frequently asked questions

What is shadow AI in the context of developer workstations?

Shadow AI refers to AI assets like skills, MCP servers, and agents that are stored locally on developer machines and have not been detected or registered as approved artifacts.

How does JFrog AI Security detect shadow AI on local machines?

It extends detection beyond repositories to laptops and workstations, identifying AI assets and assessing their risk and policy compliance without affecting developer productivity.

Why are local AI assets considered a security risk?

These assets can bypass standard security checks and central controls, allowing unvetted files, scripts, and plugins to run autonomously on developer machines without visibility or review.