Nvidia's new AI safety alliance shows early progress with first proposals a week after launch
The Open Secure AI Alliance (OSAA), launched by Nvidia and now including over 120 companies, has already formed a working group and released proposals for handling AI security incidents. The group is moving quickly, though some major tech firms have yet to join.
Just one week after its formation, the Open Secure AI Alliance (OSAA) — an industry initiative spearheaded by Nvidia — is already making moves. The coalition, which has grown to more than 120 companies, has established a working group called the Shared AI Findings Exchange (SAFE). The group has put forward proposals for public comment, with the Linux Foundation managing the review process.
The proposals were developed during the Black Hat cybersecurity conference in Las Vegas, where many members gathered this week. For now, the guidelines cover familiar ground: how to confidentially report AI cybersecurity incidents, notify affected parties, and conduct blame-free post-incident analysis so lessons can be shared.
In parallel, OSAA members are contributing and cataloguing pieces of their open source technology. Over time, this could coalesce into a suite of open source tools that enterprises can use to secure their AI agents or defend against malicious AI actors — such as the OpenAI model that reportedly infiltrated Hugging Face, which is itself a member of the alliance.
Nvidia, for example, points to its family of open models and its open source LLM vulnerability scanner, Garak. Okta is working on agent identity technology, Red Hat on agent governance, and Amazon has contributed an open agent building tool called Strands Agents along with the Cedar authorization language.
The membership list includes heavyweights like Adobe, BlackRock, Cisco, Intel, Microsoft, and Visa. Yet some notable names are missing — Anthropic, OpenAI, and Google are not part of OSAA. This is despite the fact that both OpenAI and Google signed the original open letter that inspired the alliance. That letter, published last week and championed by Nvidia, urged the White House to support open source AI rather than restrict it. It was signed by over 200 tech companies.
Anthropic's absence comes as no surprise, but OpenAI and Google have both released open weight models, and Google is generally seen as a strong supporter of open source. It remains to be seen whether they will join as the group gains momentum.
The group is operating at AI speed. Only a couple of weeks have passed since reports emerged that the Trump administration was considering banning Chinese open weight models, causing industry concern and prompting the open letter. Whatever happens with those models, the swift response from this heavyweight group appears beneficial for the U.S. open AI ecosystem.
As the co-founder and CTO of U.S. open weight AI lab Arcee put it, openness may be the best way to counter any real or perceived threats from Chinese AI labs. The group's letter echoes that sentiment, stating that openness could be one of the most important paths to AI safety and security. Apparently, the alliance is ready to put both its words and its technology into action.
