Cloudsmith launches ML model registry to boost security & control
Cloudsmith has launched a machine learning (ML) model registry intended to provide enterprise governance and security to ML models and datasets used in modern software development.
Organisations deploying ML technologies are increasingly concerned about the proliferation of models, compliance gaps, and potential security risks posed by public repositories. Documented cases of backdoored models being uploaded to platforms like Hugging Face and GitHub have highlighted the risk of malicious AI/ML components entering production systems, especially in the absence of automated controls.
The Cloudsmith ML Model Registry is designed to address these concerns by allowing organisations to apply consistent policies and oversight to ML models and datasets. The registry introduces built-in validation and security checks designed to ensure only safe, compliant assets are introduced to production codebases. These measures align with governance processes already commonplace for managing software packages and containers.
Registry features
The registry integrates directly with the Hugging Face Hub and software development kit (SDK). This makes it possible for development teams to use existing tools for pushing, pulling, and managing models and datasets while gaining visibility and control from a single platform. Organisations can also proxy and cache public models and datasets from Hugging Face into the Cloudsmith registry, and apply enterprise policies prior to any further use in development or deployment environments.
Introducing the new registry, Alison Sickelka, Vice President of Product at Cloudsmith, stated,
"The rapid adoption of AI/ML is transforming the kinds of software enterprises are building, but most organizations still lack the governance to manage models and datasets safely. With this launch, we're bringing the same enterprise-grade controls, traceability, and security to AI/ML assets that Cloudsmith customers rely on for every other part of their software supply chain."
The features supported in the new registry include the ability to centralise the management of models, containers, and language-specific packages in a unified repository. It offers compatibility with the Hugging Face SDK and ecosystem, enabling workflow continuity for developers. Organisations can proxy and cache open source models from Hugging Face, thereby enforcing their own security and compliance policies prior to use.
The platform also supports secure delivery by surfacing security, compliance, and quality indicators in its Enterprise Policy Management dashboard. Assets can be automatically quarantined, blocked, or approved according to policy. Integration with existing continuous integration/continuous delivery (CI/CD) pipelines for model validation and deployment is supported, as well as fine-grained access controls, entitlement tokens, and audit trails to protect proprietary assets. The repository structure within Cloudsmith remains flexible, supporting organisation by project, environment, or customer delivery requirements.
Managing risk
The increasing reliance on open source and third-party models exposes organisations to new types of supply chain risk. Allowing ML models into critical environments without appropriate validation can lead to compromised systems, data leaks, or breaches of compliance obligations. By offering built-in checks and policy enforcement, Cloudsmith aims to mitigate these risks.
With the ML Model Registry, teams are able to track the full lifecycle of AI and ML models - from development and training, through to validation, deployment, and ongoing maintenance. This oversight supports organisational efforts to ensure the integrity, compliance, and ongoing performance of their models at each stage of use.
Early access to the ML Model Registry is now available to organisations developing with machine learning models and datasets and seeking to introduce enterprise-level governance and security into their workflows.
