JFrog is expanding AI Security with Local Shadow AI Detection. This feature detects AI assets directly on developers’ workstations. The company is also introducing AI Asset Scanning with Bring Your Own LLM Key, which lets companies verify skills, plugins, and APM packages using their own language model.
With this move, JFrog extends AI supply chain security to areas that pipeline and registry checks cannot reach. The new capability builds on Shadow AI Detection, a feature JFrog previously added to its platform to detect unapproved internal models and external AI APIs.
JFrog says the rationale is clear. Developers install skills, plugins, MCP servers, and hooks via Git, Slack, marketplaces, or simply a ZIP file. This often happens outside security teams’ purview. Each of these assets runs using the developer’s permissions and credentials. Who installed what, and did anyone authorize it? That is far from always clear.
“Developers are already running AI agents organizations can’t see. JFrog AI Security brings them under control,” says CEO Shlomi Ben Haim. He says the question is no longer whether agents are entering the supply chain, but whether organizations can show what they’ve affected.
Local scanning, with a proprietary LLM
Local Shadow AI Detection is deployed via the JFrog Agent Plugin, using mobile device management (MDM), or through the agent harness settings. The tool identifies skills, plugins, and MCP servers, regardless of which repository or marketplace they come from. The scan is deterministic and runs on the workstation itself. Only metadata and a risk assessment are sent back to the JFrog Platform. There, it compares each asset to source data to clarify what has been approved.
AI Asset Scanning with BYOK performs semantic scans through the customer’s own LLM account. Prompts and skills go directly from the JFrog Platform to the customer’s model endpoint, without a JFrog gateway in between. Costs are billed through the customer’s own provider. For self-hosted customers, BYOK is actually the only way to use AI Asset Scanning.
These features are available immediately to users with JFrog Unified Security and Ultimate Security. At launch, BYOK supports Anthropic Opus 4.8, OpenAI GPT 5.6 Sol, and Google Gemini 3.1 Pro.