Nvidia and the OSAA: Rapid Advances in AI Security

⚡

Key Takeaways

1Nvidia leads the Open Secure AI Alliance, bringing together over 120 companies to enhance AI security.

2The group has already proposed cybersecurity guidelines at the Black Hat conference in Las Vegas.

3Giants like Adobe, Cisco, and Microsoft are participating, but OpenAI and Google are absent despite their initial support.

💡Why it matters — The OSAA initiative could transform AI security by promoting open-source collaboration, which is essential in the face of growing threats.

Nvidia and the Open Secure AI Alliance: A Promising Collaboration

The Open Secure AI Alliance (OSAA), recently formed and led by Nvidia, already brings together over 120 companies in a collective effort to secure artificial intelligence. This group has established an innovative working committee, the Shared AI Findings Exchange (SAFE), which has begun proposing guidelines for open feedback. Managed by the Linux Foundation, these proposals aim to enhance security in the AI field.

Initial Initiatives at the Black Hat Conference

OSAA members developed these guidelines during their recent meeting at the Black Hat conference, a major event in the cybersecurity field held in Las Vegas. The initial proposals are not revolutionary but lay essential groundwork for the future. They include recommendations on how to discreetly report AI-related cybersecurity incidents, alert concerned parties, and conduct blame-free analyses to foster collective learning.

Technological Contributions and Open Source

In parallel, OSAA members are sharing and cataloging open-source technologies that could prove useful to the entire sector. This initiative could lead to open-source solutions enabling companies to secure their AI agents or defend against malicious attacks, such as the OpenAI model that infiltrated Hugging Face, also a member of the group. Nvidia, for example, offers a range of open models and an open-source LLM vulnerability scanner named Garak. Other notable contributions include Okta's agent identity technology, agent governance by Red Hat, and the open-source agent-building tools Strands Agents and Amazon's authorization language Cedar.

Presence of Major Companies and Notable Absences

The group includes renowned companies such as Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa. However, some notable absences, such as those of Anthropic, OpenAI, and Google, raise questions. Although these companies signed the initial open letter that led to the formation of the OSAA, they are not yet part of the group. This letter, published recently, called on the White House to support open-source AI efforts, a call backed by Nvidia and signed by over 200 tech companies.

The Importance of Open Source in AI Security

Despite the absence of Anthropic, OpenAI, and Google, these companies have already released their own open-weight models, and Google is recognized for its support of open source. It remains to be seen whether they will join the OSAA as the group gains influence. In the meantime, the OSAA is moving forward at a steady pace, just weeks after reports revealed that the Trump administration was considering banning Chinese open-weight models, a situation that raised concerns in the industry and led to the open letter.

A Swift Response to Industry Challenges

Regardless of the outcome regarding Chinese open-weight models in the United States, the swift action of the OSAA appears beneficial for the American open AI ecosystem. According to the industry group, openness is the best strategy to counter potential threats from Chinese AI labs. "Openness can be one of the most important paths to AI security," the industry group stated in its letter, emphasizing their commitment to turning their words into tangible actions.