Software supply chain company JFrog Ltd. today announced a new machine learning lifecycle integration between JFrog Artifactory and MLflow, an open-source software platform originally developed by Databricks Inc.

The new integration is designed to give JFrog users a way to build, manage and deliver machine learning models and generative artificial intelligence-powered apps along with other software development components in a streamlined, end-to-end DevSecOps workflow. Through the integration, companies can validate the security and provenance of machine learning models to ensure responsible AI practices.

The integration is seeking to address issues wherein 80% or more of machine learning models built to create new AI-powered applications fail to deploy because of technical issues with integrating models into existing operations. JFrog’s integration with MLflow helps organizations overcome the issue by uniting MLflow’s model development solution with existing, mature DevOps workflows for end-to-end visibility, automation, control and traceability.

“Part of helping organizations scale their ability to embrace and successfully deliver AI- and gen AI-powered applications is to allow developers and data science teams to manage models with trust, the same way they manage other packages,” said Yoav Landman, chief technology officer at JFrog. “This includes using a universal, scalable, single system of record for all binaries, with the ability to control versioning, apply security checks and control models’ lifecycle.”

Building on previous successful integrations between JFrog, Amazon SageMaker and Qwak AI Ltd., the combination of JFrog Artifactory and MLflow gives machine learning engineers and Python, Java and R developers the freedom to work with their preferred tool stack, using Artifactory as their model registry.

In addition, JFrog’s platform proxies Hugging Face, allowing developers to access available open-source models easily while also looking out for malicious models and enforcing license compliance. The solution also comes with the software security features and scanners provided by the JFrog Platform to maintain risk-free machine learning applications.

The ability to detect malicious models is also a key feature at a time when they’re starting to rapidly grow in number. In February, the JFrog Security Research team discovered hundreds of malicious AI models in the Hugging Face AI repository that pose a significant risk of data breaches or attacks.

Photo: JFrog