Red Hat Launches Asago: The Open-Source AI Governance Framework Automating Policy-to-Production Controls

#asago #redhat #aigovernance #opensource

Nvidia, Microsoft, and academic partners." />

Red Hat announced Asago on August 4, 2026, an open-source framework that automates the translation of AI governance policies into production-ready operational controls. The project, released under the Apache License 2.0, combines a four-stage workflow with mappings to the NIST AI Risk Management Framework, OWASP Top 10 for LLMs, and the EU AI Act, and generates deployment artifacts for Kubernetes, Terraform, and Ansible.

Asago stands for AI Safety and Governance Orchestration. Red Hat, an IBM subsidiary, is leading the effort alongside a partner group that includes IBM Research, Nvidia, Microsoft, MIT Lincoln Laboratory, the Alan Turing Institute, and universities in Austria and North Carolina. The project builds on prior work done by Red Hat and Nvidia through the Open Secure AI Alliance. It is currently in its project formation phase, with a public GitHub repository and community governance participation already open.

Enterprise adoption of AI systems has moved past experimental prototypes into production environments where models run for months or years. That shift has exposed a gap that Asago aims to close. Compliance teams write policy documents. AI engineers run safety tests. Infrastructure operators configure deployment environments. These three groups have traditionally worked in parallel, connecting their work through manual processes that take months. Asago automates that connection.

Key Takeaways

Asago = AI Safety and Governance Orchestration, announced by Red Hat on August 4, 2026
Apache License 2.0 open-source project currently in formation phase
Four-stage workflow: Understand policies, Test AI systems, Select controls, Deploy configurations
Maps requirements to NIST AI RMF, OWASP Top 10 for LLMs, and EU AI Act using the IBM AI Risk Atlas
Generates Kubernetes, Terraform, and Ansible artifacts for hybrid cloud and on-premises environments
Partners include IBM Research, Nvidia, Microsoft, MIT Lincoln Laboratory, Alan Turing Institute
Preserves end-to-end audit trail linking each deployed control back to its originating policy clause

What Asago Actually Does

The core problem Asago targets is the manual workflow that enterprises currently use to convert governance policies into safely deployed AI systems. A compliance officer writes a policy document. An AI engineer has to interpret that document, map its requirements to relevant safety frameworks, run tests against the model, select appropriate guardrails, and configure the infrastructure to enforce them. Each step involves different tools, different teams, and different output formats. The result is a process that can take months and is prone to errors as requirements move between groups.

Asago replaces that manual chain with a standardized four-stage workflow. The first stage, Understand, interprets the organization's governance policies and maps their requirements to established frameworks. It uses the IBM AI Risk Atlas as its mapping engine, connecting policy clauses to the NIST AI Risk Management Framework, the OWASP Top 10 for Large Language Model Applications, and the European Union AI Act. The second stage, Test, generates and runs safety tests tailored to the specific risks of the AI system being evaluated. The third stage, Select, recommends mitigations and guardrails based on test results, recording the rationale for each recommendation to support auditor review. The fourth stage, Deploy, translates the selected controls into deployment-ready configurations for hybrid cloud and Kubernetes environments, producing Kubernetes manifests, Terraform infrastructure-as-code, and Ansible playbooks.

The entire workflow preserves an end-to-end audit trail. Every deployed control can be traced back to the specific policy clause that required it, along with the test evidence and rationale that justified it. That traceability is the difference between a governance process that an auditor can verify and one that relies on manual documentation.

If you are interested in how other open-source projects are tackling AI safety, Darkmoon's approach to autonomous AI penetration testing covers a complementary angle in the same security space.

Why the Shift to Production AI Changes the Governance Problem

Generative AI tools became widely available in 2023. By 2024, enterprises were running pilot programs and proof-of-concept projects. The governance approach for those early deployments was relatively simple: set up a review process, run a few safety checks, and deploy with monitoring. Most organizations did not need infrastructure automation. They were not running autonomous agents that operated without human oversight for extended periods.

The situation changed as AI systems moved into production. Autonomous agents now run for days or weeks, making decisions and taking actions without constant human supervision. Multi-agent systems coordinate across teams, applications, and infrastructure. The governance requirements for these systems are more detailed and more numerous than for simple chatbots. A single policy document might contain hundreds of requirements spanning data privacy, model safety, access control, and audit compliance.

Red Hat stated in its announcement that the asago initiative is needed specifically because of this shift. Enterprises are moving beyond experimental AI projects toward systems and autonomous agents that remain in production for long periods. The gap between broad governance requirements and the specific tests, guardrails, and infrastructure configurations that engineers need has grown wider. Manual translation of policy into production controls does not scale to that gap.

The Partner Group and Its Lineage

Asago brings together a diverse set of organizations. Red Hat leads the project. The founding partners include IBM Research, which contributes the AI Risk Atlas and governance expertise, and Nvidia, which brings infrastructure and AI system experience. Microsoft provides enterprise cloud and governance tooling perspective. MIT Lincoln Laboratory contributes cybersecurity research capabilities. The Alan Turing Institute, the UK's national institute for data science and AI, adds academic and policy research. Alquimia AI, Brave Software, and the EvalEval coalition contribute specialized AI safety and evaluation expertise. Interdisciplinary Transformation University Austria and North Carolina State University provide academic research partnerships.

The project builds on work done through the Open Secure AI Alliance, a prior initiative between Red Hat and Nvidia. That alliance focused on securing AI supply chains and deployment infrastructure. Asago extends that work into the governance and compliance layer, automating the translation of policy requirements into technical controls.

For more on how major infrastructure players are approaching AI governance, the managed agent infrastructure stack article covers how OpenComputer, Nebius, and Anthropic are building deployment platforms that will need governance integration.

How Asago Fits Into Existing Governance Frameworks

Asago does not create a new governance framework. It maps existing frameworks to production controls. The three primary frameworks it targets are the NIST AI Risk Management Framework, the OWASP Top 10 for Large Language Model Applications, and the EU AI Act. The IBM AI Risk Atlas acts as the mapping engine that connects policy requirements in an organization's documents to specific controls within these frameworks.

The NIST AI RMF provides a risk management structure for AI systems. The OWASP Top 10 for LLMs identifies the most critical security risks specific to large language model applications. The EU AI Act is a regulatory framework that classifies AI systems by risk level and imposes different requirements based on that classification. Asago's workflow connects an organization's internal policy document, which may reference one or all of these frameworks, to the specific technical controls needed in a production deployment.

The project also aligns with broader open-source governance initiatives. The OneCLI credential gateway project addresses a related problem: securing the secrets and credentials that AI agents use in production. Asago and OneCLI operate at different layers of the same stack. Asago determines what controls are needed based on policy. OneCLI manages the credentials those controls depend on.

What the Four-Stage Workflow Looks Like in Practice

The four stages of the Asago workflow form a continuous pipeline from policy document to deployed system. Here is how each stage functions in practice.

The Understand stage ingests governance policy documents. These can be PDFs, DOCX files, or structured policy representations. The system interprets the policy text and extracts specific requirements. It then maps each requirement to the appropriate framework control using the IBM AI Risk Atlas. The output is a structured requirements traceability matrix linking policy clauses to NIST, OWASP, and EU AI Act controls.

The Test stage uses the requirements matrix to generate safety tests. These tests are tailored to the specific risks of the AI system under evaluation. For a customer service chatbot, the tests might focus on hallucination detection and sensitive data leakage. For a medical diagnosis assistant, they might emphasize accuracy benchmarks and adverse outcome testing. The system runs these tests against the model and collects results.

The Select stage analyzes test results and recommends mitigations. If a test reveals a hallucination risk, the system might recommend a guardrail that validates responses against a knowledge base. If a test reveals data leakage, it might recommend input filtering and output sanitization controls. Each recommendation includes a rationale that explains why the control addresses the specific risk. This rationale is recorded for auditor review.

The Deploy stage translates the selected controls into deployment artifacts. For Kubernetes environments, it generates deployment manifests and configuration objects. For infrastructure-as-code workflows, it produces Terraform files that define the necessary resources. For configuration management, it creates Ansible playbooks that enforce the controls. The output is production-ready infrastructure that implements the governance requirements.

The audit trail connects all four stages. An auditor can start at a deployed control in a Kubernetes manifest and trace it back through the Terraform configuration, the Ansible playbook, the selected mitigation, the test result, the mapped framework control, and finally to the originating policy clause. That traceability is what makes the process auditable rather than opaque.

Current Status and How to Participate

Asago is in its project formation phase. The GitHub repository is public. Community governance participation is open. Developers, researchers, and enterprise early adopters can view the repository, review the project roadmap, and participate in governance discussions through the project website at asago.ai.

The project website describes the workflow, the partner group, and the goals and anti-goals of the initiative. The anti-goals section is notable: Asago is not building a new governance framework. It is not replacing human judgment. It is not a closed-source commercial product. The project positions itself as infrastructure for governance automation, not governance itself.

For more on how other open-source projects are giving AI agents operational capabilities, Screenpipe's persistent memory approach shows another layer of the agent infrastructure stack that governance frameworks like Asago will need to account for.

Conclusion

Asago represents a practical approach to a problem that is becoming more pressing as AI systems move into production. The gap between governance policy and operational controls has been bridged manually, with significant time and error rates. Automating that translation while preserving audit traceability addresses a real enterprise need.

The partner group is substantial. Red Hat, IBM Research, Nvidia, Microsoft, MIT Lincoln Laboratory, and academic institutions bring complementary expertise in governance, security, infrastructure, and AI research. The Apache 2.0 license and open governance model give the project the structure needed for enterprise adoption.

The project is still in formation. The four-stage workflow, framework mappings, and artifact generation capabilities are described in the project documentation but may evolve as the community develops them. What is clear is that the problem Asago targets is real and growing. Enterprises moving AI systems into production need governance that is both rigorous and efficient. Asago aims to provide that through automation rather than manual processes.

FAQ

Q: What does Asago stand for?

A: Asago is an acronym for AI Safety and Governance Orchestration. The name reflects the project's focus on automating the connection between AI safety policy and governance enforcement in production environments.

Q: How is Asago different from existing AI governance tools?

A: Most existing tools focus on a single stage of the governance process, such as policy interpretation or safety testing. Asago automates the full pipeline from policy document to deployed control, with traceability linking each stage to the next. It also generates production infrastructure artifacts rather than just reports or recommendations.

Q: What governance frameworks does Asago support?

A: Asago currently maps to the NIST AI Risk Management Framework, the OWASP Top 10 for Large Language Model Applications, and the European Union AI Act. The IBM AI Risk Atlas acts as the mapping engine that connects organizational policy requirements to these frameworks.

Q: What deployment outputs does Asago generate?

A: Asago generates Kubernetes manifests, Terraform infrastructure-as-code files, and Ansible playbooks. These artifacts are production-ready and can be deployed to hybrid cloud or on-premises environments. The system preserves an audit trail linking each artifact back to its originating policy clause.

Q: Is Asago ready for production use?

A: Asago is currently in its project formation phase. The GitHub repository is public and community governance participation is open, but the project is still being developed. Organizations interested in early access can review the repository and participate in governance discussions through the asago.ai website.