Managing open-weight model risk in enterprise deployments

Microsoft and NVIDIA are publishing Risk Management for Open-Weight Models in Enterprise, the first in a planned series of joint resources for enterprises weighing open-weight deployments.

• The paper argues that weight openness alone does not determine an AI system's risk, and that risk instead depends on the model, use case, deployment, and safeguards.
• It introduces a shared taxonomy that sorts risks into five categories: Cybersecurity, Data Protection, Content, Human-AI Interaction, and Bias.
• The guidance is grounded in established standards and frameworks, including ISO standards and the NIST AI Risk Management Framework, and is meant to be adapted by enterprises to their own systems, use cases, and risk criteria.
• The paper maps evaluations to a selection of risks across the five categories so teams can start from a risk that matters for their deployment and find evaluations that produce relevant evidence.
• Microsoft Foundry is cited as offering built-in and custom evaluations, automated red teaming, and configurable content-safety controls, which teams should validate for their model and deployment, including preview status.
• This first paper focuses on the model layer, and Microsoft and NVIDIA say future work will extend the approach across the runtime, harness, and application layers.