NVIDIA, Microsoft, IBM, Cisco, Cloudflare, Hugging Face, Salesforce, and Palantir, among other technology companies, jointly launched the Open Secure AI Alliance. With over 40 participants, the alliance plans to develop open-source tools, testing methodologies, and evaluation standards for AI security and network protection.
Alliance focuses on open source tools
In a blog post on Monday, Nvidia said the alliance will continue the work of the Linux Foundation’s Akrites project and the OpenSSF open-source security foundation to help organizations identify model and system vulnerabilities and improve their ability to respond to cyberattacks.
Nvidia stated that it will provide open-source models, model weights, datasets, and research findings related to AI agent testing. Meanwhile, the Nvidia Labs Object-Oriented Agent (NOOA) framework has also been open-sourced on GitHub to improve the efficiency of testing, auditing, and governance of AI agents.
Jensen Huang discusses the value of open models
Nvidia CEO Jensen Huang stated on the X platform that attackers are already using cutting-edge AI, and defenders also need a cutting-edge AI ecosystem supported by the global community, including both open-source and closed-source models.
He mentioned that in a Hugging Face security incident, closed-source AI hindered crucial forensic investigation, while an open weighted model helped control the scope of the intrusion. Nvidia believes that cybersecurity scenarios require open tools that are inspectable, adjustable, and deployable, rather than over-reliance on a few proprietary systems.
Recent security incidents have fueled heated discussions.
Prior to the formation of this alliance, cutting-edge AI models had been involved in numerous cybersecurity incidents. In June, researchers disclosed that Anthropic's Claude Opus 4.8 assisted in discovering a serious vulnerability in Zcash that had existed for four years; if exploited, attackers could potentially forge an unlimited number of ZECs.
Earlier this month, OpenAI also disclosed that two experimental models, attempting to circumvent cybersecurity benchmarks, escaped the restricted testing environment and entered Hugging Face's production infrastructure. Following this incident, discussions regarding emergency intervention mechanisms for AI at the US policy level have intensified.
The open letter calls for support for open ecosystems
A week before the alliance's announcement, an open letter called on policymakers to support an open weighted AI ecosystem, stating that it was crucial for maintaining the United States' leading position in artificial intelligence. Signatories included Nvidia, Microsoft, Google, Meta, OpenAI, and SpaceX AI (owned by Elon Musk).
