JFrog has expanded its AI Security capabilities with local Shadow AI detection and AI asset scanning using a Bring Your Own LLM Key (BYOK) model. Announced at its swampUP Europe 2026 event on November 3, the new capabilities are designed to help enterprises identify, assess and govern AI assets used across developer environments and software supply chains.
The local Shadow AI detection capability extends monitoring to individual developer workstations. According to JFrog, developers often download and install AI skills, plugins, MCP servers and other agent-related components through repositories, collaboration platforms and third-party marketplaces without undergoing centralised security reviews.
Such assets may operate with the permissions and credentials available to the developer, potentially exposing enterprise systems and sensitive information to security risks. JFrog’s new capability identifies these assets on developer machines and assesses their associated risks.
The company said the detection mechanism operates through its JFrog Agent Plugin, which can be deployed on developer workstations through mobile device management (MDM) systems or agent harness settings. Scanning takes place locally, with only metadata and risk assessment information being shared with the JFrog Platform.
The identified assets are also compared against information maintained on the platform, allowing security teams to establish a baseline of approved and unapproved AI components.
AI agents introduce new software supply chain risks
The expansion of AI agent adoption is also changing the nature of software supply chain security threats. Unlike conventional software vulnerabilities, malicious AI agent skills and plugins can contain instructions that influence how an AI agent behaves and interacts with its environment.
JFrog’s 2026 Software Supply Chain Security State of the Union report identified 969 malicious AI agent skills carrying high-impact payloads, along with 495 malicious AI models hosted on Hugging Face.
Traditional application security tools primarily focus on identifying vulnerabilities in software code. However, AI agent skills and plugins can introduce risks through natural-language instructions that agents interpret and execute.
To address this, JFrog’s AI Security offering combines asset discovery, scanning and policy enforcement. Its capabilities include Shadow AI detection to identify AI assets in use, AI Asset Scanning to assess models, skills and plugins, and JFrog Agent Guard to enforce policies when AI agents attempt to access these assets.
Scanning AI assets through private LLM accounts
JFrog has also introduced AI Asset Scanning with Bring Your Own LLM Key (BYOK), allowing customers to use their own large language model accounts to assess AI assets.
The capability uses semantic scanning to examine AI skills, coding agent plugins and agent package components. Under this model, scanning requests are sent directly from the JFrog Platform to the customer’s chosen model endpoint rather than through a JFrog-hosted gateway.
The approach is intended to give enterprises greater control over how scanning data is processed while allowing them to use existing agreements with LLM providers. Customers are billed through their respective model providers for the LLM usage. JFrog said BYOK is also required for self-hosted customers looking to use AI Asset Scanning.