New JFrog Security Tools Detect Malicious AI Assets
The platform now scans developer workstations for risky AI plugins and models to secure internal software supply chains.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Do you trust the automated AI plugins and agent software currently running on your own computer?
JFrog has introduced Local Shadow AI Detection and AI Asset Scanning for its security platform to identify unapproved AI plugins and model assets. These tools allow security teams to perform deterministic scanning on developer workstations and route model verification through private LLM keys.
Why it matters
As developers increasingly integrate AI agents without central oversight, security teams lack the visibility to identify malicious instructions hidden within agent skills and plugins. JFrog's update addresses this gap by enabling local visibility into assets that traditional Application Security tools often overlook.
The 2026 Software Supply Chain Security State of the Union report identified 969 malicious AI agent skills and 495 malicious models on Hugging Face. These figures highlight the volume of threats facing the 6,600 organizations currently using the JFrog platform.
The players
JFrog
A software company focused on binary repository management and supply chain security for 6,600 organizations worldwide.
Anthropic
A developer of AI models, including the Opus 4.8 architecture now supported by JFrog.
OpenAI
An AI research lab and provider of the GPT 5.6 Sol model.
A technology company providing the Gemini 3.1 Pro model for integration with security scanning tools.
The details
Local Shadow AI Detection functions by deploying an agent plugin via mobile device management or existing harness settings to catalog skills, plugins, and Model Context Protocol (MCP) servers on workstations. AI Asset Scanning utilizes semantic analysis—a method of evaluating the meaning of code or instructions—by routing prompts directly to a customer's private LLM account. This 'Bring Your Own Key' (BYOK) approach verifies the safety of assets against customer-defined security policies without exposing internal data to third-party model providers.
Timeline
October 21, 2026: JFrog announced the availability of the new AI security capabilities.
December 31, 2025: The end of the fiscal year for the JFrog 10-K report.
The Tech Race
This release follows the 2026 Software Supply Chain Security State of the Union report, which identified nearly 1,500 malicious AI assets. JFrog is positioning its platform to compete against standard AppSec tools that lack specific capabilities for auditing agent-based AI workflows.
These capabilities are now available to users with JFrog Unified Security and Ultimate Security subscriptions. Organizations can now enforce security standards for AI agents directly on developer workstations without abandoning private model keys.
The takeaway
The security industry is shifting toward workstation-level auditing to neutralize malicious AI agent instructions before they enter the code pipeline. Organizations should verify their current JFrog tier to ensure access to these local scanning capabilities.
Further reading
For broader trends in enterprise-grade security tools for LLMs, visit the Artificial Intelligence section.
More information
To review technical requirements and feature specifications, visit the JFrog AI Security product information page.
Live Poll
Do you trust the automated AI plugins and agent software currently running on your own computer?
