On October 16, 2026, Microsoft and NVIDIA published a joint whitepaper titled Risk Management for Open-Weight Models in Enterprise to help organizations address deployment risks.
Shared Risk Taxonomy
Microsoft said the paper introduces a shared risk taxonomy and an approach to managing model risks grounded in established standards, including International Organization for Standardization (ISO) standards and the National Institute of Standards and Technology (NIST) AI Risk Management Framework. The companies stated that weight openness alone does not determine an AI system's risk, which instead depends on the model, use case, deployment, and safeguards.
According to Microsoft, the initial taxonomy organizes risks into five broad categories: Cybersecurity, Data Protection, Content, Human-AI Interaction, and Bias. The framework was developed jointly by Andrew Berkley and April Chen of Microsoft alongside Michael Boone, Eileen Long, and Arun Rao of NVIDIA.
Deployment Context and Evaluations
Microsoft stated that open- and closed-weight models share underlying risks, but how those risks materialize depends on the deployment context. The first paper focuses primarily on model-layer risks, though teams must also assess permissions, data access, tools, runtime, and operational controls.
To help teams turn the taxonomy into practice, the paper maps evaluations to risks across the five categories. Microsoft noted that its Microsoft Foundry platform helps enterprises assess quality, groundedness, and safety using built-in and custom evaluations, automated red teaming, and configurable content-safety controls.
What to do
- Enterprises should adapt the risk taxonomy to their own systems, use cases, and risk criteria, connecting relevant risks to evaluations, mitigations, and accountable owners.
- Validate control availability for model deployments and combine platform capabilities like Microsoft Foundry evaluations with application-specific testing, monitoring, and human oversight.
Key facts and where they come from
- Microsoft and NVIDIA published Risk Management for Open-Weight Models in Enterprise on October 16, 2026.
Today, Microsoft and NVIDIA are publishing Risk Management for Open-Weight Models in Enterprise
- The initial taxonomy organizes risks into five categories: Cybersecurity, Data Protection, Content, Human-AI Interaction, and Bias.
organizes risks into five broad categories: Cybersecurity, Data Protection, Content, Human-AI Interaction, and Bias.
- The risk management approach is grounded in ISO standards and the NIST AI Risk Management Framework.
grounded in established standards and frameworks, including International Organization for Standardization (ISO) standards and the National Institute of Standards and Technology (NIST) AI Risk Management Framework.
