Nutanix: Open Infrastructure Is the Foundation of Secure AI

The recent formation of the Open Secure AI Alliance by NVIDIA and leading tech pioneers underscores a truth we've long championed at Nutanix: true enterprise AI safety and security should not exist in a black box. The defenses that protect our critical infrastructure must be built on open agents, models, open harnesses, and transparent tools that defenders can study, adapt, and deploy on their own terms.

Spearheaded by NVIDIA, the Linux Foundation, and the OpenSSF, this alliance represents a major movement to develop and share open-source technologies that safeguard software and agents in the age of AI. Nutanix is proud to be part of the newest cohort of industry leaders joining the Open Secure AI Alliance, as announced this week.

As the industry pushes for a shared, open ecosystem for AI defense, Nutanix is actively accelerating this mission. Our teams are shaping the open-source AI landscape across MLOps, inference, AI routing, and fundamental safety benchmarks, backed by peer-reviewed research and core open-source maintenance.

Setting the Standard with MLCommons & Research Leadership

Safety and performance require rigorous, standardized testing. Nutanix plays a critical role in developing the benchmarks that the industry relies on today:

• Core AI Safety Research: Members of our team have collaborated on the foundational MLCommons AI Safety research papers, e.g.  "Pluralis v0.1: Towards a Multicultural, Multimodal, Multilingual Benchmark for AI Risk and Reliability". As core members of AI safety, reliability, and security standardization within MLCommons, Nutanix supports safety and risk evaluation standards for generative AI systems across complex hazard categories. Our technical contributions encompass safety data curation, ensemble evaluation (including LLM-as-a-judge via Judge-Pluralis), jailbreak and failure mode analysis, and multi-modal, multi-regional, and multilingual evaluations.

• MedPerf: We co-created MedPerf to enable privacy-preserving, federated evaluation for medical AI models directly on local healthcare data without compromising sensitive patient records. The landmark framework behind this platform was published in the Nature Machine Intelligence paper, "Federated benchmarking of medical artificial intelligence with MedPerf".

• MLPerf Storage: We co-created the MLPerf Storage benchmark and served as the inaugural co-chair to standardize performance measurement and address storage I/O bottlenecks across intensive, large-scale AI workloads.

The AI Stack Needs a Security Layer

Beyond benchmarking and published research, our engineering teams are directly building the open-source infrastructure required to run models safely at scale:

• KServe: As key maintainers, we are advancing KServe as the cloud-native standard for highly scalable, multi-framework model serving on Kubernetes. KServe helps organizations deploy open and fine-tuned models efficiently and securely across hybrid cloud environments.

• Envoy AI Gateway: Also, as key maintainers, we are delivering enterprise-grade AI traffic management through the Envoy AI Gateway. We're building robust Model Context Protocol (MCP) routing, fine-grained authorization, LLM-specific telemetry, and dynamic load balancing for complex inference workloads.

A Shared Vision for Open AI Security

Our inclusion in the Open Secure AI Alliance aligns perfectly with our work alongside strategic partners like NVIDIA, Dell Technologies, AMD, Intel, and Cisco, among others. Real AI safety depends on the entire agent stack - from the underlying hybrid cloud infrastructure to the models and the routing in between. By maintaining key pieces of open-source infrastructure and co-creating rigorous industry standards, Nutanix is committed to supporting the open, observable, and secure foundation necessary to mobilize the global community of AI defenders.

Sources

Open Source Maintainers 

• https://github.com/envoyproxy/ai-gateway/blob/main/MAINTAINERS.md

• https://github.com/kserve/community/blob/main/MAINTAINERS.md

Key Research Papers 

• Medical Perf benchmarks: https://www.nature.com/articles/s42256-023-00652-2   

• AI Safety https://arxiv.org/abs/2607.06196

• AI Safety Benchmarks https://arxiv.org/abs/2503.05731

• Additional Research https://rajatghosh.me/