Broadcom joins Nvidia AI security alliance: 'We haven't forgotten open source'

Broadcom has joined Nvidia's Open Secure AI Alliance, expanding the venture's membership from nearly 40 founders to 74, with Broadcom's Infrastructure Software Group leading its involvement. The alliance aims to secure enterprise AI systems through open-source solutions, countering reliance on closed AI models. Broadcom's Chris Wolf cited the company's history in open-source leadership, including contributions to Kubernetes and projects like Harbor, Antrea, Velero, and Contour.

Broadcom became the latest firm to join the Open Secure AI Alliance from Nvidia, with Broadcom's software infra division to lead on the venture’s open-source mission to secure enterprise AI systems. The VMware owner joined Netskope, Nutanix, Akamai, and others late last week in membership, expanding the venture’s initial count of almost 40 founders to 74. Inaugural members include Cisco, Nokia, Palo Alto Networks, Microsoft, SK Telecom, and Hewlett Packard Enterprise. With its Open Secure AI Alliance, Nvidia is pushing for an open defensive stack around AI agents, and pushing back against cybersecurity dependence on a small number of closed and centrally controlled AI models. While Nvidia’s ever-updating membership https://blogs.nvidia.com/blog/open-secure-ai-alliance/ lists VMware by Broadcom, a Broadcom exec clarified the firm’s Infrastructure Software Group is leading the silicon giant’s involvement in the venture. “Broadcom has historically played significant leadership roles in driving open and interoperable technology ecosystems, and we've been among the top contributors to Kubernetes for the past decade, with projects including Harbor, Antrea, Velero, and Contour originating from our engineering teams,” said Chris Wolf, global head of AI and advanced services for VMware at Broadcom. Wolf added Broadcom is the steward and primary maintainer of widely adopted open source software outside of VMware, including Java-based application maker Spring and application-to-AI connector Spring AI, as well as business messaging backbone RabbitMQ. As revealed in April, Broadcom also served as one of the original members of Anthropic’s Project Glasswing security venture. https://www.sdxcentral.com/news/anthropic-bug-hunter-launched-with-all-star-cast-and-good-old-us-patriotism/ “In the months since that project was announced, we have continued collaborating with industry peers, enterprise organizations, and government entities to share findings and best practices. The fast-moving security landscape presented a natural opportunity to bring that expertise and track record of driving meaningful community outcomes to the Open Secure AI Alliance,” Wolf told SDxCentral , while adding virtualization defines how Broadcom approaches AI safeguarding. As he put it, virtual machines VMs have long been the primary mechanism for isolating runtimes between tenants and across security boundaries, on-premises, and in the public cloud. As such, when models, agents, and services run in isolated, policy-governed environments, the blast radius of any failure or compromise is contained by design. Wolf pointed to VMware Cloud Foundation VCF https://www.sdxcentral.com/news/vmware-introduces-cyber-compliance-tool-and-agentic-ai-updates/ AI security updates around isolation, encryption, access control, threat scanning, and traffic governance, as seen across Avi Load Balancer, Tanzu Platform, and vDefend. Open source versus open weight AI Nvidia’s Open Secure AI Alliance came in response to a recent Hugging Face breach https://www.sdxcentral.com/news/palo-alto-networks-ceo-warns-openai-hugging-face-breach-brings-browser-sase-to-the-fore/ where the AI firm was hacked by closed-source agents from OpenAI. According to Nvidia, the incident delivered a clear reminder that “cyber defenders need open, frontier agentic systems for self-defense. “When closed AI tools – unable to distinguish attackers from defenders – blocked essential forensic analysis, Hugging Face ran the open-weight GLM 5.2 model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion,” Nvidia explained https://blogs.nvidia.com/blog/open-secure-ai-alliance/ . Shortly after, the chip giant, along with Cisco https://www.sdxcentral.com/news/cisco-security-ai-aces-openai-but-anthropic-thats-a-mythos-mystery/ , OpenAI, and Hugging Face, would co-sign a policy statement supporting open-weight AI models. Open-weight models hand over model files or weights to developers minus the training data, full code, and methodology that would come in a fully open source release. Somewhat controversially, Anthropic abstained from co-signing the statement of support, arguing it was too sweeping https://www.anthropic.com/news/position-open-weights-models and that powerful, cybersecurity-focused models need to be closed off to prevent misuse by bad actors, hence the exclusive nature of its Project Glasswing https://www.sdxcentral.com/news/anthropic-expands-glasswing-granting-150-new-firms-access-to-powerful-mythos-model/ . Anthropic is also not a member of the Nvidia alliance, and neither are Amazon and Google although the latter is represented by its Wiz subsidiary https://www.sdxcentral.com/news/google-cloud-closes-wiz-acquisition-begins-platform-player-brawl/ . Oracle remains the only hyperscaler to have neither signed the letter nor joined Nvidia’s venture. SDxCentral reached out to Nvidia for comment on the benefits of an open-weight approach compared to a fully open-source one, but did not receive a response. In Broadcom's view, the difference is defined by enterprise-ready maturity, especially when it comes to security. “Linus's Law states that ‘given enough eyeballs, all bugs are shallow,’” Wolf explained. “In cybersecurity, trust must be earned through verification, and open source models and harnesses provide exactly that: global visibility and auditability, community-based testing, and distributed innovation. That is the foundation of why open models matter for security, and why Broadcom supports them.” The Broadcom exec said open-weight models give organizations the ability to run AI on their own terms, adding: “You can take a capable, publicly available model, fine-tune it to your industry or use case, and run it in an environment you fully control, with your data never leaving your perimeter. That combination of performance, flexibility, and data sovereignty explains why open-weight models have become central to private AI deployments, and why they have proven effective at reducing costs for use cases like vulnerability discovery.” “The alliance's current emphasis on open-weight models reflects where enterprise adoption is concentrated today, not a judgment that open source is less valuable. As open source models mature, the scope of that conversation will follow. Broadcom supports commercial, open source, and open-weight alternatives because that is what our customers are asking for,” Wolf said.