Red Hat has launched Asago, an open-source community project designed to translate AI governance policies directly into production-ready deployment code. The initiative arrives as the EU AI Act takes effect and enterprises face mounting pressure to demonstrate compliance without slowing AI deployment to a crawl. Backed by a roster that includes NVIDIA, IBM, Microsoft, Brave Software, MIT Lincoln Laboratory, North Carolina State University, and the Alan Turing Institute, Asago aims to replace manual compliance reviews with an automated, auditable workflow that spans risk mapping, assessment, mitigation, and runtime orchestration across hybrid cloud and Kubernetes environments.
The project builds on work Red Hat and NVIDIA began inside the Open Secure AI Alliance, a cross-industry group formed after a high-profile security incident at Hugging Face exposed the limits of closed-model tooling for incident response. Asago is released under the Apache License 2.0 and remains in its formation phase, with a public GitHub repository open to developers, academic researchers, and enterprise early adopters. Red Hat positions the framework as infrastructure rather than a one-off certification tool, emphasizing continuous audit trails that stay checkable as autonomous agents run in production.
What's New / Specs
Asago operates across four stages that move from policy text to running controls. The first stage, risk mapping, ingests an organization's governance policy and maps its requirements against established frameworks including the NIST AI Risk Management Framework, the OWASP LLM Top 10, and the EU AI Act as catalogued through IBM's AI Risk Atlas. This step converts policy language into a machine-readable risk profile automatically, eliminating the manual cross-referencing that compliance teams typically perform.
The second stage, risk assessment, generates and executes scenarios tailored to the specific use case. Rather than running a generic checklist, the system probes for the harmful behaviors that the risk mapping identified. The third stage, risk mitigation, recommends guardrails based on what testing surfaced and builds a rationale trail designed to withstand regulatory scrutiny. The final stage orchestrates those recommended controls into deployment-ready configurations for hybrid cloud and Kubernetes environments, producing declarative artifacts for Kubernetes, Terraform, and Ansible that Red Hat says are infrastructure-agnostic.
- License: Apache 2.0
- Frameworks referenced: NIST AI RMF, OWASP LLM Top 10, EU AI Act (via IBM AI Risk Atlas)
- Output formats: Declarative configs for Kubernetes, Terraform, Ansible
- Target environments: Hybrid cloud, Kubernetes
- Current phase: Formation, public GitHub repository
- Founding contributors: Red Hat, NVIDIA, IBM, Microsoft, Brave Software, MIT Lincoln Laboratory, NC State University, Alan Turing Institute, EvalEval coalition, IT:U (Austria), Alquimia AI (partner)
Red Hat's stated goal is to compress deployment timelines from months to days. Steven Huels, Red Hat's VP of AI Engineering, frames Asago as the next logical step after the company's Lightwell initiative, which focuses on securing the open-source supply chain from AI-driven vulnerabilities. Stuart Battersby, Red Hat's AI safety and model evaluation architect, describes the project as a true collaborative endeavor bringing together technology industry, academia, and government stakeholders, and encourages additional collaborators from global jurisdictions to ensure maximum coverage of AI safety viewpoints.
Why It Matters
The EU AI Act's entry into force creates a concrete deadline for enterprises operating in or serving European markets. Organizations must now demonstrate that their AI systems meet risk-tiered requirements for transparency, human oversight, data governance, and robustness. Manual compliance processes that worked for a handful of models become untenable as companies deploy fleets of autonomous agents across hybrid environments. Asago's approach of generating a continuous audit trail — where each policy clause ties to a specific test and each test ties to a runtime control — addresses a core regulatory demand: traceability from principle to practice.
The project's infrastructure-agnostic outputs matter for enterprises that run workloads across multiple clouds and on-premises clusters. A safety posture defined once as declarative configuration can be applied consistently without re-engineering for each target environment. This reduces the configuration drift that often undermines compliance programs. The involvement of major cloud and infrastructure players — Microsoft, IBM, NVIDIA, Red Hat — signals that the industry recognizes governance as a shared infrastructure layer rather than a competitive differentiator.
Asago also connects to the broader Open Secure AI Alliance narrative. That alliance formed after Hugging Face suffered a security breach that its closed AI tooling could not help contain; the company switched to the open-weight GLM 5.2 model to analyze more than 17,000 actions taken during the intrusion. Alliance members are contributing complementary pieces of the agent governance stack: NVIDIA's NOOA project targets the harness layer that constrains what models can call; HPE contributes to SPIFFE/SPIRE for cryptographic workload identity; Hugging Face contributed Safetensors to the PyTorch Foundation to address remote code execution risks in model file formats; Microsoft contributes MDASH, a multi-agent scanning harness for code vulnerability detection; and IBM and Red Hat's Lightwell project applies digitally-signed patches across the open-source supply chain. Asago sits at the policy-to-code translation layer of this stack.
Our Take
Asago represents a credible attempt to industrialize AI governance, but the gap between formation-phase code and production-grade compliance tooling is substantial. The project's value proposition hinges on two unproven claims: that automated risk mapping can reliably capture the nuance of regulatory text across jurisdictions, and that the generated guardrails will satisfy auditors without extensive customization. The founding contributor list is impressive, but multi-stakeholder open-source projects often struggle with governance disputes once implementation details surface — particularly around risk-mapping standards where regulatory interpretation diverges.
The Lightwell connection is notable. IBM and Red Hat committed $5 billion to Project Lightwell in May 2026, backed by more than 20,000 engineers and early adopters including Bank of America, BNY, Citi, Goldman Sachs, JPMorgan Chase, Mastercard, Morgan Stanley, Royal Bank of Canada, State Street, Visa, and Wells Fargo. Lightwell establishes a trusted enterprise clearinghouse for open-source software security, using AI-driven validation and testing across massive code volumes. Asago's policy-to-code automation could become a consumption layer for Lightwell's validated patches and supply-chain assurances, but that integration remains speculative.
Enterprises evaluating Asago should treat it as a framework to contribute to and shape, not a product to adopt. The absence of deployed customer case studies, benchmark data under live regulatory audit, and a defined dispute-resolution mechanism for contributor disagreements means the project carries significant adoption risk. Organizations with mature MLOps platforms may find more immediate value in extending their existing policy-as-code tooling rather than waiting for Asago to mature. However, the project's infrastructure-agnostic output format and alignment with the EU AI Act's structural requirements make it worth tracking for teams building long-term governance architecture.
FAQ
What is Asago and who is behind it?
Asago is an open-source community project launched by Red Hat that automates the translation of AI governance policies into production-ready deployment configurations. Founding contributors include NVIDIA, IBM, Microsoft, Brave Software, MIT Lincoln Laboratory, North Carolina State University, the Alan Turing Institute, the EvalEval coalition, Austria's Interdisciplinary Transformation University (IT:U), and Alquimia AI as a partner. The project is released under the Apache License 2.0.
How does Asago's four-stage workflow operate?
The workflow begins with risk mapping, where uploaded governance policies are mapped against the NIST AI RMF, OWASP LLM Top 10, and EU AI Act via IBM's AI Risk Atlas. Risk assessment then generates and runs use-case-specific scenarios to probe for flagged harmful behaviors. Risk mitigation recommends guardrails based on test results and builds a rationale trail. Finally, orchestration converts those controls into declarative configurations for Kubernetes, Terraform, and Ansible across hybrid cloud environments.
What problem does Asago solve for enterprises facing the EU AI Act?
Asago addresses the compliance bottleneck created when organizations must govern growing fleets of autonomous agents under the EU AI Act's risk-tiered requirements. By generating a continuous audit trail linking each policy clause to a specific test and runtime control, it provides the traceability regulators demand. The infrastructure-agnostic output format allows a single safety posture to be applied consistently across multiple clouds and on-premises clusters without re-engineering.
Is Asago production-ready today?
No. The project is in its formation phase with a public GitHub repository. Red Hat's announcement includes no deployed customer case studies, no benchmark data showing the "days not months" deployment claim under live regulatory audit, and no defined mechanism for resolving disputes between contributors over risk-mapping standards. Enterprises should view it as a framework to contribute to and shape rather than a product to adopt immediately.
How does Asago relate to the Open Secure AI Alliance and Project Lightwell?
Asago builds on work Red Hat and NVIDIA began inside the Open Secure AI Alliance, which formed after Hugging Face used the open-weight GLM 5.2 model to analyze 17,000+ intrusion actions during a breach that closed tooling could not contain. Alliance members contribute complementary governance stack layers: NVIDIA's NOOA for agent harnesses, HPE's SPIFFE/SPIRE for workload identity, Microsoft's MDASH for code scanning, and IBM/Red Hat's Lightwell for supply-chain patching. Lightwell is a $5 billion commitment backed by 20,000+ engineers and major financial institution early adopters.
