Nvidia shows early progress in AI group, a week after launch

A week after its launch, the Open Secure AI Alliance (OSAA), the industry group spearheaded by Nvidia, is already moving at a rapid clip. The consortium, which has swelled to more than 120 member companies, has formed a working group with a memorable moniker: the Shared AI Findings Exchange, or SAFE. The group has already pushed out proposals for public comment, with the Linux Foundation, a participating member, handling the administrative side of that process.

These initial drafts took shape while members convened at the Black Hat conference in Las Vegas this week, the cybersecurity world’s annual gathering spot. The early guidelines are modest in scope, but they lay critical groundwork. They outline protocols for confidentially reporting AI cybersecurity incidents, notifying affected parties, and conducting blame-free post-mortems so the broader community can learn from what went wrong.

Beyond the paperwork, OSAA members are also busy contributing and cataloging pieces of their open source technology stacks that could prove useful to the collective effort. If history is any guide for industry alliances like this, those contributions may eventually crystallize into an open source framework that enterprises can use to lock down their AI agents or defend against rogue AI threats, such as the OpenAI model that managed to infiltrate Hugging Face, which is itself a member of this group.

The roster of contributions is already impressive. Nvidia points to its family of open models and an open source LLM vulnerability scanner called Garak. Okta is working on agent identity technology, Red Hat is focusing on agent governance, and Amazon has thrown in both an open agent building tool, Strands Agents, and an authorization language called Cedar. Those are just a few examples from a long list.

The alliance has attracted a who’s who of the tech world, including Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa. But the absence of some major players is hard to ignore. Anthropic, OpenAI, and Google have all declined to join so far.

That’s a curious development, especially since both OpenAI and Google signed the original open letter that gave birth to this alliance. That letter, published last week and backed by more than 200 tech companies, urged the White House to champion open source AI efforts rather than stifle them. Nvidia was the driving force behind it.

Anthropic’s reluctance to sign on or join the group is hardly shocking, given its history. But OpenAI and Google are a different story. Both have released open weight models of their own, and Google has long positioned itself as a friend of open source. It remains to be seen whether they’ll come around as the OSAA gains momentum.

The speed of all this is noteworthy. It has only been a couple of weeks since reports surfaced that the Trump administration was weighing a ban on Chinese open weight models, a move that sent ripples of concern through the industry and ultimately led to the open letter. The OSAA’s rapid formation and early output suggest a sense of urgency.

Whatever the U. S. decides to do about Chinese open weight models, the swift action from this heavyweight coalition looks like a positive sign for the domestic open AI ecosystem. Some insiders, like the co-founder and CTO of U. S. open weight AI lab Arcee, argue that openness is the most effective answer to any threat, real or perceived, from Chinese AI labs.

“Openness may be one of the most important paths to AI safety and security,” the group wrote in its letter. Based on the pace of their work so far, they appear ready to back that statement up with action, not just words.