Devsecops company JFrog on September 13 introduced ML Model Management, a set of capabilities for the JFrog Software Supply Chain Platform designed to streamline the management and security of machine learning models.
Using ML Model Management and the JFrog Software Supply Chain Platform, organisations can manage their proprietary models in Artifactory and proxy the Hugging Face model repository to cache the open source AI models they rely on, bringing them closer to production and development and protected from deletion or modification.
Further, new ML security capabilities in JFrog Xray allow companies to detect and block malicious models and models with licenses that don’t comply with company policies. Users also can store home-grown or internally augmented ML models with access controls and versioning history. ML models can be bundled and distributed as part of any software release.
JFrog said increasing numbers of organisations are incorporating ML models into applications. With government regulations requiring software vendors to list what is inside their software, JFrog believes it will not be long before these guidelines are extended to include AI and ML models. The integration of the ML model management capability gives customers a way to store, secure, and manage ML models alongside other software components, the company said.
JFrog on September 13 also unveiled these other capabilities for JFrog Software Supply Chain Platform:
- Static Application Security Testing (SAST), to help with scanning source code for zero-day security vulnerabilities. JFrog SAST helps minimise false positives and prioritise remediation efforts via contextual analysis.
- Open-Source Software (OSS) Catalog, providing a search engine for software packages in the JFrog UI or via an API, backed by both public and JFrog data, and giving users insight into security and risk metadata associated with OSS packages. The catalog is featured as part of the JFrog Curation component of Software Supply Chain Platform.
- Release Lifecycle Management (RLM), creating an immutable release bundle defining a software package and its components early in the software development lifecycle. RLM also leverages anti-tampering systems, compliance checks, and evidence capture to collect data on each release bundle.
JFrog said that to address the rise in software supply chain attacks, securing at the binary level with immutable software bundles was a “must” because this was the only way to certify what was being released, and certifying the release was the only way to certify safety for use.