JFrog launches local AI security for developer laptops

JFrog has launched new AI security features to identify and govern AI tools installed on developer workstations, extending its AI security offering beyond pipeline and registry controls.

The update adds local detection of AI skills, plugins and Model Context Protocol servers on developer laptops, along with scanning of AI assets through a customer's own large language model account. The additions are intended to give security teams visibility into AI assets installed outside central oversight.

The move comes as businesses adopt AI agents in software development while facing new risks from tools and instructions that sit outside traditional application security processes. JFrog said these assets can arrive through repositories, marketplaces, collaboration platforms and compressed files, and may run with a developer's existing permissions.

Research cited by JFrog found Australia leads globally in automated controls to block unapproved developer extensions and MCP servers, with 47% of organisations using such measures. But, it said, those controls do not necessarily cover AI assets installed directly on local machines before they reach standard security checks.

Local oversight

JFrog said its local shadow AI detection is delivered through its agent plugin on the developer workstation, where it can identify AI-related assets regardless of source. Scanning runs on the local device and sends only metadata and a risk assessment back to the platform, according to the company.

The approach is intended to help organisations compare what is installed on a device with approved assets already recorded in the central platform. JFrog said this creates a baseline security teams can use for policy enforcement across development and production environments.

JFrog also introduced AI asset scanning with what it calls Bring Your Own LLM Key, allowing customers to use their own model accounts to assess skills, coding agent plugins and APM packages. Prompts and skills are sent directly from the JFrog platform to the customer's chosen model endpoint rather than through a JFrog-hosted gateway, the company said.

The feature is likely to appeal to regulated organisations seeking tighter control over where data is processed. JFrog said self-hosted customers must use Bring Your Own LLM Key to run AI asset scanning.

Changing threat model

JFrog linked the launch to findings in its 2026 Software Supply Chain Security State of the Union report, which identified 969 malicious AI agent skills carrying what it described as high-impact payloads, alongside 495 malicious AI models on Hugging Face. The company said these assets differ from conventional software risks because they often rely on plain-language instructions that an AI agent reads and follows.

That marks a shift in software supply chain security, which has traditionally focused on vulnerable or malicious code. Security teams now need to assess not just executable code, but also the intent embedded in prompts, plugins, skills and models that direct agent behaviour.

"Developers are already running AI agents organisations can't see. JFrog AI Security brings them under control," said Shlomi Ben Haim, Co-Founder and Chief Executive Officer at JFrog. "Until now, the conversation around AI agents has focused on what they can build vs. what they can break or who answers for it when that happens. The question is no longer whether agents will touch your supply chain - it's whether you can prove what they touched. Our customers live in a zero-trust world and can't trust AI agents or developers using AI blindly. JFrog brings detection, scanning, and policy enforcement for the AI-powered software supply chain into one solution."

JFrog said the product combines three elements: local detection of AI assets, scanning of models and tools, and policy enforcement when an agent attempts to use them. It positions this as a single record of AI assets for development and security teams.

Support for Bring Your Own LLM Key initially covers Anthropic's Opus 4.8, OpenAI's GPT 5.6 Sol and Google's Gemini 3.1 Pro. The new functions are available to customers on JFrog Unified Security and Ultimate Security plans.

Yuval Fernbach, Vice President and CTO, JFrog AI Security, said the company was responding to a visibility gap at the start of the software development process. "Security teams need visibility that starts at the developer's machine - before an asset ever reaches a coding agent or is shared through Artifactory. Regulated customers need scanning that points at their own models," Fernbach said. "Local Shadow AI Detection and AI Asset Scanner address both needs in a single solution: JFrog AI Security."