CodeCraft: Red Hat, NVIDIA, IBM Collaborate on Transforming AI Policy into Action

Red Hat Launches asago Project for AI Governance

Red Hat has introduced the asago project, an open-source initiative aimed at transforming AI governance policies into deployable code. This project seeks to streamline the process of turning policies into actionable deployment strategies for AI technologies.

The asago project is described as an automated workflow that bridges the gap between engineering and compliance teams by simplifying the complex steps involved in deploying AI technologies. With the recent implementation of regulations like the EU AI Act, organizations are faced with the choice of either slowing down AI innovation through manual reviews or risking ungoverned AI agents running in production without proper oversight.

Building on the collaboration between Red Hat and NVIDIA within the Open Secure AI Alliance, the asago project is released under the Apache License 2.0. Currently in its formation phase, the project has a repository open on GitHub for developers, academic researchers, and enterprise early adopters to contribute to governance.

The workflow outlined by Red Hat consists of four key stages. The first stage involves risk mapping, where the organization’s governance policy is analyzed and matched against established frameworks such as the NIST AI RMF, the OWASP LLM Top 10, and the EU AI Act. This mapping process automatically generates a risk profile based on policy language, eliminating the need for manual cross-referencing by compliance teams.

Following risk mapping, the project moves into risk assessment, generating tailored scenarios to identify potential harmful behaviors flagged during the mapping phase. Risk mitigation comes next, with the system recommending guardrails based on the identified risks and creating a rationale trail for review.

Asago then orchestrates the recommended controls into deployment-ready configurations for hybrid cloud and Kubernetes environments, streamlining the process and reducing deployment timelines from months to days, according to Red Hat.

One of the key features of the project is the focus on creating a continuous audit trail throughout the entire process. Each policy clause is linked to a specific test, which in turn is tied to a runtime control. This level of traceability allows reviewers to track each active control in a live deployment back to the policy line that justified it.

Steven Huels, Red Hat’s VP of AI Engineering, emphasizes the importance of establishing clear operational guardrails as organizations transition from experimental AI pilots to long-running autonomous agents. He sees the asago project as a critical infrastructure requirement for enterprise AI.

The project has garnered contributions from a diverse group of stakeholders beyond Red Hat and NVIDIA, including Brave Software, IBM Research, Microsoft, MIT Lincoln Laboratory, North Carolina State University, and The Alan Turing Institute. The collaborative nature of the project aims to address a wide range of AI safety and security challenges.

Asago’s outputs are designed to be infrastructure-agnostic, providing declarative configurations for Kubernetes, Terraform, and Ansible, ensuring a consistent safety posture across different cloud environments.

While the project is still in its early stages and has not been production-tested, it represents a significant step towards automating AI governance processes and enhancing AI safety practices. The asago project invites developers, researchers, and enterprise teams to join the community-driven effort and contribute to the evolution of AI governance practices.

In conclusion, the asago project by Red Hat is a collaborative initiative that aims to streamline AI governance processes and enhance AI safety practices through automation and continuous auditability. With contributions from a diverse group of stakeholders, the project represents a significant step towards establishing clear operational guardrails for enterprise AI deployments.