JFrog extends DevSecOps playbook to AI governance

Software security specialist JFrog is extending its platform to manage artificial intelligence (AI) models with the same rigour as software artefacts, moving to unify DevSecOps, machine learning operations (MLOps) and governance into a single platform.

In doing so, the company is leveraging its position as a central registry for software artefacts to bring order to the emerging field of AI development, according to Sunny Rao, JFrog’s senior vice-president for Asia-Pacific.

“AI models are nothing but analogous to software,” he said in an interview with Computer Weekly in Singapore. “We’re already maintaining the system of record as a registry for all software artefacts, so it’s only logical that we are the system of records for all AI models.”

This move addresses a challenge for organisations that have mature DevSecOps practices for software but are running AI projects in separate, less-governed environments. “All of the old practices that we had fixed with DevSecOps were creeping into AI,” said Rao. “Trying to bring the same DevSecOps methodology to AIOps became very important.”

Central to JFrog’s efforts to manage AI models is the introduction of machine learning bills of materials (ML-BOM), which is akin to a traditional software bill of materials (SBOM), an inventory of components and software dependencies in software applications that has become a standard in software security. Rao explained that an ML-BOM must account for two distinct layers of provenance. “One is the model itself, and the second is the data sets used to train the model,” he said.

While JFrog can validate a model’s integrity and immutability, the data set introduces complex vectors like privacy concerns, licensing and bias. “How did you source this data? How much bias did you introduce? These concepts are enshrined into our ML-BOM,” he said, adding that JFrog incorporates governance frameworks like Singapore’s fairness, ethics, accountability and transparency principles, with digital signatures at every stage to ensure a clear audit trail.

This capability extends to closed-source models where data provenance is unknown. “If a particular AI model comes in with certain restrictions, or you don’t know the provenance of the data, we will flag it to you,” said Rao. This allows highly regulated industries to make risk-based decisions on whether to use such models.

The JFrog platform also tackles the risk of transitive models – where one AI model calls another – by enabling organisations to define policies that can block or flag models that violate their governance standards before they even enter the development environment.

Rao observed that most customers in the Asia-Pacific region are currently focused on laying their DevSecOps foundations to prepare for the “inevitable flow of agentic AI capabilities”.

“We’re seeing a flurry of activity and deployments in the SecOps space in order to get ready for what is coming down the line,” he said, referring to agentic AI adoption, which is currently concentrated in areas such as AI-assisted coding, and data analysis in the financial services and retail sectors.

To help organisations operationalise their agentic AI deployments, JFrog acquired Qwak AI (now JFrog ML) in 2024, delivering capabilities such as real-time monitoring of model performance, A/B testing and model experimentation, as well as cost benchmarking.

Rao noted that JFrog’s aim is to provide a single, open platform that integrates with the entire software development ecosystem, from integrated development environments and source code repositories like GitHub to model registries such as Hugging Face and Nvidia inference microservices. This is designed to counter the “massive proliferation of point solutions” that creates blind spots and inefficiencies for organisations.

JFrog plans to unveil further details on its platform integrations and partnerships at its upcoming SwampUp user conference in Napa, California from 8–10 September.

In Asia-Pacific, which has more developers than the rest of the world combined, Rao said many companies have less legacy infrastructure compared with North America and Europe, allowing them to adopt modern DevSecOps and MLOps more quickly and efficiently.

To meet this demand, JFrog has expanded its presence in the region, with India now its second-largest development centre outside Israel, where the company was founded.